commit a5c319a1fcdb8635683b4ba1950900d7708aab59
Author: dimon <dinlo-irk@yandex.ru>
Date:   Sat Jun 13 16:36:27 2026 +0800

    Initial commit: Ideogram 4 Prompt Builder
    
    PyQt6 desktop app for building Ideogram 4 JSON captions: bbox canvas,
    palette editor, presets, prompt library with previews, localisation (en/ru),
    light/dark themes, and ComfyUI dependency check + generation.
    
    Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..4b3848f
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,15 @@
+# Python
+__pycache__/
+*.py[cod]
+
+# Generated at runtime / regenerated from code on first launch
+translations.json
+
+# User-specific and runtime state (not part of the application source)
+comfy_settings.json
+draft.json
+prompt_library.json
+prompt_previews/
+
+# Editor / agent
+.claude/
diff --git a/README.md b/README.md
new file mode 100644
index 0000000..905a738
--- /dev/null
+++ b/README.md
@@ -0,0 +1,159 @@
+# Ideogram 4 Prompt Builder
+
+**English** · [Русский](#ideogram-4-prompt-builder-ru)
+
+A desktop GUI (PyQt6) for building structured JSON captions for **Ideogram 4** and ComfyUI workflows, with a prompt library, reference-image canvas, localisation, light/dark themes, and direct generation through a ComfyUI server.
+
+![English interface](eng-vlack.png)
+
+## Run
+
+```powershell
+python ideogram_prompt_builder.py
+```
+
+Requires `PyQt6` (no other third-party dependencies):
+
+```powershell
+pip install PyQt6
+```
+
+## What it builds
+
+Prompts follow the schema from `docs/prompting.md`:
+
+- `high_level_description`
+- `style_description` with either `photo` or `art_style`
+- `compositional_deconstruction.background`
+- `compositional_deconstruction.elements`
+- optional uppercase HEX color palettes
+- optional bounding boxes in normalized `0-1000` coordinates
+
+Actions live in a menu bar (**File / Edit / Library / ComfyUI / View**) plus a slim toolbar (Generate, Undo/Redo, Save to library, Library, Copy) and the language/theme controls on the right. The right-hand panel is tabbed: **JSON** (output + validation) and **Result** (the generated image).
+
+## Editing
+
+- Move and resize layout boxes directly with the mouse on the bbox canvas.
+- Palette fields accept comma-separated HEX, clickable swatches and a popup color picker, with a live `n/limit` counter and invalid-color highlighting.
+- **Undo / Redo** (`Ctrl+Z` / `Ctrl+Y`).
+- **Duplicate**, **reorder** (up/down) and add elements from **templates** (Character / Title text / Background object).
+- The validation list is clickable — clicking an element-specific message selects that element.
+- Text fields have a right-click translation menu (`Translate to RU` / `Translate to EN`, results cached).
+- Work is autosaved to `draft.json`; on the next launch you are offered to restore it.
+
+## Reference image & zoom
+
+In the composition panel you can load a **reference image** (file or paste from clipboard) drawn under the bbox grid; the **grid scale** slider zooms the grid and the reference scales with it.
+
+## Prompt library
+
+The **Library** menu saves the current caption (optionally with a preview image), updates the entry you loaded from, and opens the library browser, where you can:
+
+- search by name / tag / description and edit per-entry **tags**;
+- load any saved prompt back into the editor for reuse and editing;
+- attach a preview from a file or **paste it from the clipboard**, or remove it;
+- rename, delete entries, and view the preview + summary;
+- **export / import** the whole library (prompts + previews) as a single `.zip`.
+
+The library is stored in `prompt_library.json` next to the app, with preview images in `prompt_previews/` (created on first save).
+
+## ComfyUI integration
+
+The **ComfyUI** menu connects the builder to a running ComfyUI server:
+
+- **ComfyUI settings** — host, port and HTTPS, with a *Test connection* button. Stored in `comfy_settings.json`.
+- **Check ComfyUI** — verifies that every model, sampler and custom node the bundled `ideogram4NSFWComfyui_v11.json` workflow needs is installed on the server, and lists anything missing.
+- **Generate in ComfyUI** — converts the bundled workflow to API format, injects the current compact JSON caption, submits it and retrieves the generated image. The result appears in the **Result** tab and can be saved to a file or into the library.
+
+## Appearance & localisation
+
+- **Theme** (View menu) toggles a light / dark theme.
+- The interface language is switched at runtime from the **Language** selector; the default is **English**.
+
+UI strings are loaded from `translations.json`, created on first run from bundled `en` / `ru` translations. To add a language, add a top-level key with the same string keys (and optionally a display name in `LANGUAGE_NAMES`). Missing keys fall back to English then to the key name. Theme and language are saved in `comfy_settings.json`.
+
+## Compact JSON for ComfyUI
+
+The output can be copied in pretty or compact form. Compact JSON matches the recommended serialization style for inference and can be pasted into the Ideogram 4 prompt field in ComfyUI.
+
+---
+
+<a name="ideogram-4-prompt-builder-ru"></a>
+
+# Ideogram 4 Prompt Builder (RU)
+
+[English](#ideogram-4-prompt-builder) · **Русский**
+
+Десктопное GUI-приложение (PyQt6) для сборки структурированных JSON-промтов для **Ideogram 4** и ComfyUI: с библиотекой промтов, холстом с референс-изображением, локализацией, светлой/тёмной темой и прямой генерацией через сервер ComfyUI.
+
+![Русский интерфейс](ru-white.png)
+
+## Запуск
+
+```powershell
+python ideogram_prompt_builder.py
+```
+
+Нужен только `PyQt6` (других сторонних зависимостей нет):
+
+```powershell
+pip install PyQt6
+```
+
+## Что собирается
+
+Промты соответствуют схеме из `docs/prompting.md`:
+
+- `high_level_description`
+- `style_description` с одним из `photo` или `art_style`
+- `compositional_deconstruction.background`
+- `compositional_deconstruction.elements`
+- опциональные палитры HEX в верхнем регистре
+- опциональные bbox в нормализованных координатах `0-1000`
+
+Действия вынесены в меню (**Файл / Правка / Библиотека / ComfyUI / Вид**) плюс компактная панель инструментов (Сгенерировать, Отменить/Повторить, Сохранить в библиотеку, Библиотека, Копировать) и переключатели языка/темы справа. Правая панель — вкладки: **JSON** (вывод + валидация) и **Результат** (сгенерированное изображение).
+
+## Редактирование
+
+- Перемещайте и масштабируйте рамки прямо мышью на холсте bbox.
+- Поля палитры принимают HEX через запятую, кликабельные образцы и всплывающий выбор цвета, со счётчиком `n/лимит` и подсветкой некорректных цветов.
+- **Отмена / Повтор** (`Ctrl+Z` / `Ctrl+Y`).
+- **Дублирование**, **изменение порядка** (вверх/вниз) и добавление элементов из **шаблонов** (Персонаж / Заголовок / Фоновый объект).
+- Список валидации кликабельный — клик по сообщению об элементе выделяет этот элемент.
+- У текстовых полей есть контекстное меню перевода (`Перевести на RU` / `Перевести на EN`, с кэшированием).
+- Работа автосохраняется в `draft.json`; при следующем запуске предлагается восстановить черновик.
+
+## Референс-изображение и масштаб
+
+В панели композиции можно загрузить **референс-изображение** (из файла или вставить из буфера), которое рисуется под сеткой bbox; ползунок **масштаба сетки** увеличивает сетку, и референс масштабируется вместе с ней.
+
+## Библиотека промтов
+
+Меню **Библиотека** сохраняет текущий промт (по желанию с превью), обновляет загруженную запись и открывает браузер библиотеки, где можно:
+
+- искать по имени / тегам / описанию и редактировать **теги** записи;
+- загрузить любой сохранённый промт обратно в редактор для повторного использования и правки;
+- прикрепить превью из файла или **вставить из буфера обмена**, либо убрать его;
+- переименовывать, удалять записи и просматривать превью + сводку;
+- **экспортировать / импортировать** всю библиотеку (промты + превью) одним `.zip`.
+
+Библиотека хранится в `prompt_library.json` рядом с приложением, превью — в `prompt_previews/` (создаются при первом сохранении).
+
+## Интеграция с ComfyUI
+
+Меню **ComfyUI** связывает приложение с запущенным сервером ComfyUI:
+
+- **Настройки ComfyUI** — хост, порт и HTTPS, с кнопкой *Проверить соединение*. Хранятся в `comfy_settings.json`.
+- **Проверить ComfyUI** — проверяет, что все модели, семплеры и кастомные ноды, нужные встроенному workflow `ideogram4NSFWComfyui_v11.json`, установлены на сервере, и перечисляет отсутствующие.
+- **Сгенерировать в ComfyUI** — конвертирует встроенный workflow в API-формат, подставляет текущий compact JSON, отправляет запрос и получает изображение. Результат показывается во вкладке **Результат** и может быть сохранён в файл или в библиотеку.
+
+## Внешний вид и локализация
+
+- **Тема** (меню Вид) переключает светлую / тёмную тему.
+- Язык интерфейса переключается на лету через селектор **Язык**; по умолчанию — английский.
+
+Строки интерфейса берутся из `translations.json`, который создаётся при первом запуске из встроенных переводов `en` / `ru`. Чтобы добавить язык, добавьте ключ верхнего уровня с тем же набором строк (и при желании отображаемое имя в `LANGUAGE_NAMES`). Отсутствующие ключи откатываются к английскому, затем к самому ключу. Тема и язык сохраняются в `comfy_settings.json`.
+
+## Compact JSON для ComfyUI
+
+Вывод можно скопировать в pretty- или compact-виде. Compact JSON соответствует рекомендованной сериализации для инференса и вставляется в поле промта Ideogram 4 в ComfyUI.
diff --git a/docs/README.md b/docs/README.md
new file mode 100644
index 0000000..a9d347d
--- /dev/null
+++ b/docs/README.md
@@ -0,0 +1,336 @@
+<p align="center"><a href="https://ideogram.ai/" target="_blank" rel="noopener noreferrer"><picture>
+  <source media="(prefers-color-scheme: dark)" srcset="assets/ideogram_logo_darkmode.svg">
+  <source media="(prefers-color-scheme: light)" srcset="assets/ideogram_logo.svg">
+  <img src="assets/ideogram_logo.svg" alt="Ideogram" width="500">
+</picture></a></p>
+
+<p align="center"><em>Ideogram 4: Open image model at the forefront of design</em></p>
+
+<p align="center">
+  <a href="https://ideogram.ai/blog/ideogram-4.0/" target="_blank" rel="noopener noreferrer"><img src="https://img.shields.io/badge/Blog-Post-orange" alt="Blog Post"></a>
+  <a href="https://github.com/ideogram-oss/ideogram4" target="_blank" rel="noopener noreferrer"><img src="https://img.shields.io/badge/Code-GitHub-181717?logo=github" alt="Code"></a>
+  <a href="https://huggingface.co/collections/ideogram-ai/ideogram-4" target="_blank" rel="noopener noreferrer"><img src="https://img.shields.io/badge/Model-HuggingFace-blue?logo=huggingface" alt="Model"></a>
+  <a href="https://developer.ideogram.ai/" target="_blank" rel="noopener noreferrer"><img src="https://img.shields.io/badge/API-developer.ideogram.ai-purple" alt="API"></a>
+  <a href="https://ideogram.ai/" target="_blank" rel="noopener noreferrer"><img src="https://img.shields.io/badge/Official%20Site-ideogram.ai-ff69b4" alt="Official Site"></a>
+</p>
+
+<p align="center">
+  <img src="assets/samples/collage_landscape.jpg" alt="A collage of Ideogram 4 samples spanning photorealism, illustration, typography, and poster design">
+</p>
+
+
+Ideogram 4 is **[Ideogram](https://ideogram.ai)'s first open-weight text-to-image model**. It is a **state-of-the-art foundation model trained from scratch** — not a fine-tune of any existing model. It introduces a new structured JSON prompting interface, with best-in-class multilingual text rendering, deep language understanding, explicit bounding-box layout and color-palette controls, and native 2k resolution images. The easiest way to try the model is online at **[ideogram.ai](https://ideogram.ai/)**.
+
+We believe openness drives innovation, and we invite the research community to innovate with us on the forefront of visual intelligence.
+
+## Table of Contents
+
+1. [News](#news)
+2. [Model Zoo](#model-zoo)
+3. [Performance](#performance)
+4. [Quick Start](#quick-start)
+5. [Model Summary](#model-summary)
+6. [Prompting Guide](#prompting-guide)
+7. [Documentation](#documentation)
+8. [Citation](#citation)
+
+## News
+
+* **[2026-06-03]** **Ideogram 4 released!** Inference code and weights
+  are now public, and our [technical blog post](https://ideogram.ai/blog/ideogram-4.0/) is live. See the
+  [Quick Start](#quick-start) section to generate your first image, or try the
+  model online at [ideogram.ai](https://ideogram.ai/).
+
+## Model Zoo
+
+| Model | Params | Weight Quantization | Supported Hardware | Diffusers Support | License |
+| :---  | :---:  | :---:        | :---:   | :---:   | :---:   |
+| **[Ideogram 4 (nf4)](https://huggingface.co/ideogram-ai/ideogram-4-nf4)** | 9.3B | nf4 | CUDA | Yes | [Ideogram 4 Non-Commercial](model_licenses/LICENSE-IDEOGRAM-4-NON-COMMERCIAL) |
+| **[Ideogram 4 (fp8)](https://huggingface.co/ideogram-ai/ideogram-4-fp8)** | 9.3B | fp8 | All | No | [Ideogram 4 Non-Commercial](model_licenses/LICENSE-IDEOGRAM-4-NON-COMMERCIAL) |
+
+We plan to support more quantizations in the future.
+
+
+## Performance
+
+We evaluate Ideogram 4 across third-party arenas and benchmarks, standard
+open-source benchmarks, and our own internal human-preference benchmark. Across
+all of them, **Ideogram 4 is the best open-weight image model by far, and sits
+at the frontier of design.**
+
+### Design Arena
+
+[Design Arena](https://www.designarena.ai/) is a third-party image Elo
+leaderboard focused specifically on design-oriented generation. On the overall
+board, Ideogram 4 is the top-ranked open-weight model, trailing only proprietary
+GPT and Gemini models:
+
+<p align="center">
+  <img src="assets/benchmarks/design_arena.png" alt="Design Arena overall image Elo leaderboard with Ideogram 4.0 as the top open-weight model">
+</p>
+
+Filtered to open-weight models only, Ideogram 4 leads by a commanding margin,
+well ahead of the next-best open model:
+
+<p align="center">
+  <img src="assets/benchmarks/design_arena2.png" alt="Design Arena open-weight image Elo leaderboard, with Ideogram 4.0 well ahead of all other open models">
+</p>
+
+### ContraLabs
+
+[ContraLabs](https://contralabs.com/research) ran a blind typography evaluation judged by
+ten professional designers from Contra's top-earning talent. Ideogram 4 leads on
+first-place win rate, picked as the best of four models 47.9% of the time
+overall — well ahead of Gemini 3.1 Flash Image Preview (Nano Banana 2) at 30.0%,
+FLUX.2 [max] (15.5%), and Grok Imagine 1.0 (15.0%):
+
+<p align="center">
+  <img src="assets/benchmarks/contralabs_typography.png" alt="ContraLabs typography first-place win rate, with Ideogram v4 leading">
+</p>
+
+It also wins on practical usability: asked "Would you use this in real client
+work?", the same designers rated Ideogram 4 highest at 3.55 / 5 — significantly
+above Nano Banana 2 (2.84), Grok Imagine 1.0 (2.61), and FLUX.2 [max] (2.49):
+
+<p align="center">
+  <img src="assets/benchmarks/contralabs_typography2.png" alt="ContraLabs 'would you use this in real client work?' rating, with Ideogram v4 leading">
+</p>
+
+### LMArena
+
+On [LMArena](https://lmarena.ai/), a third-party text-to-image leaderboard that
+measures general-purpose text-to-image use cases, Ideogram is the top-ranked
+open-weight lab and a top-5 image generation lab overall — beaten only by giant
+companies with vastly larger budgets and resources:
+
+<p align="center">
+  <img src="assets/benchmarks/lmarena_benchmark.png" alt="LMArena text-to-image lab leaderboard with Ideogram">
+</p>
+
+### Ideogram internal eval
+
+For our internal human-preference benchmark, focused on graphic design and
+photography, we had graphic designers deeply familiar with professional design
+work do the rating blind. Bradley-Terry scores rank Ideogram 4 #2 overall —
+behind only GPT Image 2 medium — and the top open-weight model:
+
+<p align="center">
+  <img src="assets/benchmarks/ideogram_benchmark.png" alt="Ideogram internal design leaderboard with Ideogram 4.0">
+</p>
+
+### Open-source benchmarks
+
+On standard open-source benchmarks measuring core capabilities — layout control
+(7Bench), spatial reasoning and object fidelity (SpatialGenEval), text rendering
+(X-Omni OCR), and prompt alignment (Prism) — Ideogram 4 closes the gap to the
+leading closed-source models across every axis. On layout control (7Bench), it
+is significantly better than all closed-source models:
+
+<p align="center">
+  <img src="assets/benchmarks/opensource.png" alt="Five-axis capability radar comparing Ideogram 4.0 to leading closed-source models on layout control, spatial reasoning, object fidelity, prompt alignment, and text rendering">
+</p>
+
+At 9.3B parameters, Ideogram 4 delivers the best text rendering of any open-weight
+release we benchmarked — ahead of much larger models like Qwen-Image (20B),
+FLUX.2 [dev] (32B), and HunyuanImage 3.0 (80B MoE):
+
+<p align="center">
+  <img src="assets/benchmarks/opensource2.png" alt="Parameter-efficiency scatter plot showing Ideogram 4.0 at 9.3B parameters leading all other open-weight models on text rendering">
+</p>
+
+
+## Quick Start
+
+### Install
+
+```bash
+pip install .
+```
+
+If you plan to modify the code, install in editable mode instead so changes
+under `src/ideogram4/` take effect without reinstalling:
+
+```bash
+pip install -e .
+```
+
+### Model access
+
+The model weights are **gated** on Hugging Face, so you must accept the gate and
+authenticate before the code can download them — otherwise the download fails
+with a `404` / `GatedRepoError`.
+
+1. Open the model page — [ideogram-ai/ideogram-4-nf4](https://huggingface.co/ideogram-ai/ideogram-4-nf4)
+   (or [ideogram-ai/ideogram-4-fp8](https://huggingface.co/ideogram-ai/ideogram-4-fp8)) — and click
+   **Agree and access repository** to accept the license gate.
+2. Create a Hugging Face access token at
+   [huggingface.co/settings/tokens](https://huggingface.co/settings/tokens) and log in so the
+   download is authenticated:
+
+   ```bash
+   hf auth login
+   ```
+
+   Alternatively, export the token directly: `export HF_TOKEN="hf_..."`.
+
+### CLI
+
+The plain `--prompt` is rewritten into the structured JSON caption the model
+expects by a "magic prompt" LLM. By default this uses Ideogram's hosted
+magic-prompt API, which is **free** and does the expansion server-side (no local
+model or system prompt needed). It reads `IDEOGRAM_API_KEY` — get a key at
+https://developer.ideogram.ai/:
+
+```bash
+python run_inference.py \
+  --prompt "a ginger cat wearing a tiny wizard hat reading a spellbook" \
+  --output out.png \
+  --quantization "nf4" \
+  --magic-prompt-key "$IDEOGRAM_API_KEY"
+```
+
+You can also run the expansion through your own LLM provider — one of our magic-prompt
+system prompt is **open source**. See the
+[Prompting Guide](docs/prompting.md#magic-prompt) for details.
+
+For the highest-quality images, set `--height 2048 --width 2048` and
+`--sampler-preset V4_QUALITY_48`.
+
+#### Safety screening with Hive
+
+Prompt and output safety screening is performed via [Hive](https://thehive.ai/).
+Sign up and create a Text Moderation key and a Visual Content Moderation key,
+then export them as `HIVE_TEXT_MODERATION_KEY` and `HIVE_VISUAL_MODERATION_KEY`
+(or pass them via `--hive-text-key` / `--hive-visual-key`).
+
+```bash
+python run_inference.py \
+  --prompt "an isometric illustration of a tiny city floating in the clouds" \
+  --output out.png \
+  --quantization "nf4" \
+  --magic-prompt-key "$MAGIC_PROMPT_API_KEY" \
+  --hive-text-key "$HIVE_TEXT_MODERATION_KEY" \
+  --hive-visual-key "$HIVE_VISUAL_MODERATION_KEY"
+```
+
+For sampler presets, parameter reference, and optimization tips, see
+[docs/inference.md](docs/inference.md).
+
+## Model Summary
+
+Ideogram 4 is a **foundation model trained entirely from scratch**, not a
+fine-tune or distillation of any existing checkpoint. It is a flow-matching
+text-to-image model built on a **fully single-stream** Diffusion Transformer
+(DiT) architecture.
+
+**Architecture:**
+- **Fully single-stream DiT.** Text and image tokens are concatenated into one
+  unified sequence and processed through the same 34-layer transformer, with no
+  separate text or image branches. This enables deep cross-modal interaction at
+  every layer.
+- **Vision-language model as text encoder.** Instead of a text-only encoder
+  like CLIP or T5, Ideogram 4 uses
+  [Qwen3-VL-8B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct),
+  a full vision-language model that provides far richer understanding of visual
+  concepts. Hidden states are extracted from **13 intermediate layers** and
+  concatenated, giving the model multi-scale semantic features ranging from
+  surface-level token information to deep compositional understanding.
+- **Dual-branch classifier-free guidance.** The conditional (positive) and
+  unconditional (negative) branches can be independently refined, enabling
+  separate control over prompt adherence and image quality.
+- **Flexible resolution.** Native support for any resolution from 256 to 2048
+  (multiples of 16), with aspect ratios up to 6:1. A single model handles
+  everything from square thumbnails to ultrawide banners, with the noise
+  schedule auto-adjusting per resolution.
+
+**Key Capabilities:**
+- **Extreme controllability.** Ideogram 4 is trained on structured JSON
+  captions, giving users unprecedented control over composition, style,
+  lighting, color palette, typography, and spatial layout, all from a single
+  prompt.
+- **State-of-the-art text rendering.** Ideogram 4 delivers best-in-class
+  in-image text generation (signage, logos, captions, watermarks, multi-line
+  text) with high fidelity directly from the prompt.
+- **Spatial layout control.** Bounding-box coordinates in the prompt allow
+  explicit placement of subjects, text elements, and background regions.
+- **Color palette conditioning.** Specify hex colors in the prompt to steer the
+  image's dominant color scheme.
+
+For full architecture details, see
+[docs/model_architecture.md](docs/model_architecture.md). For a walkthrough of
+how the pipeline components fit together, see
+[docs/pipeline.md](docs/pipeline.md).
+
+## Prompting Guide
+
+Ideogram 4 is trained exclusively on **structured JSON captions**. While
+plain-text prompts work, you will get the best results by providing a JSON
+object that follows our caption schema.
+
+
+Key points:
+
+- **Use JSON prompts** for maximum controllability — the model was trained on
+  them and understands the structure natively.
+- **Color palette conditioning** — specify a `colour_palette` array of hex
+  colors in the style description to steer the image's color scheme.
+- **Aspect ratio flexibility** — Ideogram 4 supports a wide range of aspect
+  ratios (any multiple-of-16 resolution from 256 to 2048 on each side). This
+  is a key advantage for practical use: portraits, landscapes, banners,
+  phone wallpapers, social media formats, etc.
+- **Bounding-box layout** — specify `bbox` coordinates in the prompt to
+  explicitly place subjects, text elements, and background regions.
+- **Compositional control** — use `compositional_deconstruction` with bounding
+  boxes and per-element descriptions for precise spatial layout.
+
+
+**Why JSON-only training?** We train exclusively on JSON so that training
+and inference share a single, common prompt format. The training captions themselves are deliberately
+**extremely descriptive**: each JSON exhaustively describes everything in
+the image to maximize training efficiency. The more
+text-to-image relationships each caption pins down, the more grounded
+supervision the model extracts from a single training pair, rather than
+having to infer those relationships across many sparsely-captioned samples.
+
+**Why JSON at inference time?** Because the model was trained on captions
+that name every object explicitly, the most reliable way to get every
+requested object rendered is to mirror that pattern. Plain-text prompts still work, but
+won't perform as well since the model was only trained on structured JSON captions.
+
+**Don't want to write JSON by hand?** That's what *magic prompt* is for: it uses
+an LLM to expand a plain-text prompt into a full structured caption before
+generation, so you get JSON-quality results from a casual prompt. It runs by
+default in `run_inference.py` (see the [CLI](#cli) section).
+
+See [docs/prompting.md](docs/prompting.md) for a full guide.
+
+## Documentation
+
+| Document | Description |
+| :------- | :---------- |
+| [docs/prompting.md](docs/prompting.md) | How to write JSON prompts, color palette conditioning, aspect ratios |
+| [docs/inference.md](docs/inference.md) | Sampler presets, parameter reference, resolutions, optimization tips |
+| [docs/model_architecture.md](docs/model_architecture.md) | Architecture diagram, DiT spec, component details |
+| [docs/pipeline.md](docs/pipeline.md) | Conceptual pipeline walkthrough — how all components fit together |
+| [docs/development.md](docs/development.md) | Dev setup, pre-commit hooks, contributing |
+| [docs/safety.md](docs/safety.md) | Pre-training, post-training, and inference-time safety mitigations; how to report violations |
+
+## Citation
+
+If you find the provided code or models useful for your research, consider citing them as:
+
+
+```bibtex
+@misc{ideogram-4-2026,
+    author={Ideogram AI},
+    title={{Ideogram 4}},
+    year={2026},
+    howpublished={\url{https://ideogram.ai/blog/ideogram-4.0/}},
+}
+```
+
+## We're Hiring!
+
+We're looking for **Research Scientists** and **Research Engineers** to
+work on next-generation generative models and the products built on top of
+them. Interested candidates please apply https://jobs.ashbyhq.com/ideogram
diff --git a/docs/development.md b/docs/development.md
new file mode 100644
index 0000000..1bcad9a
--- /dev/null
+++ b/docs/development.md
@@ -0,0 +1,58 @@
+# Development
+
+## Editable install
+
+We recommend installing into an isolated environment — the dependencies include several GB of CUDA-built wheels.
+
+```bash
+python -m venv .venv && source .venv/bin/activate
+```
+
+For development, install the package in editable mode so changes to the source
+tree are picked up without reinstalling:
+
+```bash
+pip install -e .
+```
+
+or with [`uv`](https://docs.astral.sh/uv/):
+
+```bash
+uv venv && source .venv/bin/activate
+```
+
+```bash
+uv pip install -e .
+```
+
+## Pre-commit hooks
+
+This repo uses [pre-commit](https://pre-commit.com/) to run lint, format, and
+type checks (`ruff`, `mypy`, etc.) before each commit.
+
+Install once per clone:
+
+```bash
+pip install pre-commit
+pre-commit install
+```
+
+`pre-commit install` registers a git hook in `.git/hooks/pre-commit`, so it
+requires the directory to be a git repo. The hooks now run automatically on
+`git commit` against staged files.
+
+To run the hooks manually against every file in the repo (useful right after
+the first install, or in CI):
+
+```bash
+pre-commit run --all-files
+```
+
+The first run downloads each hook's environment (ruff, mypy, etc.) into
+`~/.cache/pre-commit/` and may take a minute. Subsequent runs are fast.
+
+To bump pinned hook versions in `.pre-commit-config.yaml`:
+
+```bash
+pre-commit autoupdate
+```
diff --git a/docs/inference.md b/docs/inference.md
new file mode 100644
index 0000000..0e5f6b1
--- /dev/null
+++ b/docs/inference.md
@@ -0,0 +1,63 @@
+# Inference Reference
+
+Detailed parameters, sampler presets, supported resolutions, and optimization
+tips for Ideogram 4 inference.
+
+## Sampler Presets
+
+Named presets bundle a step count, per-step CFG schedule, schedule mean (`mu`),
+and schedule standard deviation (`std`) into a single flag:
+
+```bash
+python run_inference.py \
+  --prompt "a cat wearing a tiny top hat" \
+  --sampler-preset V4_QUALITY_48 \
+  --output out.png
+```
+
+| Preset | Steps | CFG schedule | `mu` | `std` |
+| :----- | :---: | :----------- | :--: | :---: |
+| `V4_QUALITY_48` | 48 | 45 steps @ gw=7, then 3 polish steps @ gw=3 | 0.0 | 1.5 |
+| `V4_DEFAULT_20` | 20 | 18 steps @ gw=7, then 2 polish steps @ gw=3 | 0.0 | 1.75 |
+| `V4_TURBO_12` | 12 | 11 steps @ gw=7, then 1 polish step @ gw=3 | 0.5 | 1.75 |
+
+`V4_QUALITY_48` is the default. Fewer steps trade quality for speed. The full
+registry lives in
+[`ideogram4.sampler_configs.PRESETS`](../src/ideogram4/sampler_configs.py); add a
+new entry there to define your own.
+
+## Key Parameters
+
+These are the keyword arguments accepted by `Ideogram4Pipeline.__call__`. The
+defaults below apply when you call `pipe(...)` directly; `run_inference.py`
+overrides `num_steps`, `guidance_schedule`, `mu`, and `std` from the chosen
+sampler preset (see above).
+
+| Parameter | Default | Notes |
+| :-------- | :-----: | :---- |
+| `height` / `width` | 1024 | Must be multiples of 16. Supported range: 256–2048. Aspect ratios up to 6:1 or 1:6. |
+| `num_steps` | 48 | More steps = higher quality. The `V4_QUALITY_48` preset (48 steps) is a good speed/quality trade-off. |
+| `guidance_scale` | 7.0 | Constant guidance weight used when no `guidance_schedule` is given. Higher = more prompt adherence, lower = more diversity. |
+| `guidance_schedule` | `None` | Optional per-step guidance weights (loop-index order: index 0 is the final step). Overrides `guidance_scale`. |
+| `mu` | 0.5 | Logit-normal schedule mean. Auto-adjusted for resolution. |
+| `std` | 1.0 | Logit-normal schedule standard deviation. |
+| `seed` | `None` | Set for reproducible results. |
+
+## Supported Resolutions
+
+Ideogram 4 natively supports any resolution where both height and width are
+multiples of 16, within the range 256–2048 (aspect ratios up to 6:1 or 1:6).
+
+| Use case | Resolution | Aspect ratio |
+| :------- | :--------: | :----------: |
+| Square | 1024 × 1024 | 1:1 |
+| Landscape | 1536 × 1024 | 3:2 |
+| Portrait | 1024 × 1536 | 2:3 |
+| Widescreen | 1920 × 1088 | ~16:9 |
+| Ultrawide | 2048 × 768 | ~21:9 |
+| Phone wallpaper | 1024 × 1792 | ~9:16 |
+| Social banner | 1600 × 400 | 4:1 |
+
+Resolution buckets use 16-pixel increments, giving fine-grained control over
+output dimensions.
+
diff --git a/docs/model_architecture.md b/docs/model_architecture.md
new file mode 100644
index 0000000..13d8ed4
--- /dev/null
+++ b/docs/model_architecture.md
@@ -0,0 +1,45 @@
+# Model Architecture
+
+```
+prompt ─► Qwen3-VL-8B-Instruct (extract hidden states from layers (0,3,…,33,35) → concat)
+            │   
+            ▼
+    ┌──────────────────────────────────────────────────┐
+    │    Ideogram4Transformer                         │  
+    │  • 34 × Ideogram4TransformerBlock               │
+    │      – Ideogram4Attention (QK-RMSNorm, MRoPE)   │
+    │      – Ideogram4MLP (SwiGLU)                    │
+    │      – adaln scale/gate from t-embedding         │
+    │  • Ideogram4FinalLayer                          │
+    └──────────────────────────────────────────────────┘
+            │  velocity prediction
+            ▼
+    Euler flow-matching sampler with asymmetric CFG
+            │  denoised image latents
+            ▼
+    VAE decode
+            │
+            ▼
+            PIL.Image
+```
+
+The transformer is a single-stream DiT: text tokens (Qwen3-VL hidden states from
+the activation layers) and image latent tokens are concatenated into one
+sequence, modulated per-block by an AdaLN computed from the flow-matching
+timestep embedding. Attention uses QK-RMSNorm and 3D MRoPE so that text and
+image tokens share a unified positional space.
+
+Model spec:
+
+| field             | value         |
+|-------------------|---------------|
+| `emb_dim`         | 4608          |
+| `num_layers`      | 34            |
+| `num_heads`       | 18            |
+| `intermediate`    | 12288         |
+| `adanln_dim`      | 512           |
+| `rope_theta`      | 5_000_000     |
+| `mrope_section`   | (24, 20, 20)  |
+| latent channels   | 32 × 2² = 128 |
+| max text tokens   | 2048          |
+| sampler           | Euler flow-matching, logit-normal schedule, asymmetric CFG |
diff --git a/docs/pipeline.md b/docs/pipeline.md
new file mode 100644
index 0000000..87263f9
--- /dev/null
+++ b/docs/pipeline.md
@@ -0,0 +1,183 @@
+# Pipeline: How All the Components Work Together
+
+This document explains the end-to-end Ideogram 4 inference pipeline
+conceptually. For the architecture spec and code pointers, see
+[model_architecture.md](model_architecture.md).
+
+## Overview
+
+Ideogram 4 is a **flow-matching text-to-image model** built on a
+**single-stream DiT** (Diffusion Transformer). The pipeline has four main
+components:
+
+```
+ ┌─────────────┐   ┌──────────────────────┐   ┌──────────────┐   ┌───────────┐
+ │  Qwen3-VL   │   │  Ideogram4          │   │  KL VAE      │   │           │
+ │  Text       ├──►│  Transformer (DiT)   ├──►│  VAE         ├──►│  Image    │
+ │  Encoder    │   │  + Euler Sampler     │   │  Decoder     │   │           │
+ └─────────────┘   └──────────────────────┘   └──────────────┘   └───────────┘
+     frozen              trainable                 frozen
+```
+
+## 1. Text Encoder — Qwen3-VL-8B-Instruct
+
+The text encoder is a frozen [Qwen3-VL-8B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct)
+vision-language model, used in text-only mode (no vision inputs).
+
+**What it does:**
+- Tokenizes the prompt using the Qwen3 chat template.
+- Runs a forward pass through the 36-layer transformer.
+- **Extracts hidden states** from 13 specific layers: 0, 3, 6, 9, 12, 15, 18, 21,
+  24, 27, 30, 33, 35.
+- Concatenates these hidden states along the feature dimension, producing a
+  multi-scale text representation.
+
+**Why multi-layer extraction?** Different layers capture different levels of
+abstraction — early layers encode surface-level token information, while later
+layers encode deeper semantic meaning. Concatenating them gives the DiT access
+to the full spectrum.
+
+**Output:** A tensor of shape `(batch, num_text_tokens, hidden_dim * 13)`.
+
+## 2. DiT Backbone — Ideogram4Transformer
+
+The core generative model is a 34-layer single-stream Diffusion Transformer.
+
+### Sequence layout
+
+Text tokens and image latent tokens are concatenated into one sequence and
+processed through the same self-attention layers.
+
+```
+Sequence layout (per sample):
+
+  ┌───────────────────┬────────────────────────┐
+  │  text tokens      │  image latent tokens   │
+  │  (up to 2048)     │  (grid_h × grid_w)     │
+  └───────────────────┴────────────────────────┘
+           ▲                    ▲
+     Qwen3-VL features    noisy latents z_t
+```
+
+### Key components per block
+
+- **Self-attention** with QK-RMSNorm and 3D Multimodal RoPE (MRoPE). The
+  positional encoding is 3-dimensional: for text tokens it uses a 1D position
+  broadcast to 3 axes; for image tokens it uses (temporal, height, width)
+  coordinates. This lets text and image tokens coexist in a unified positional
+  space.
+- **SwiGLU MLP** — the feed-forward layer uses a gated linear unit with SiLU
+  activation.
+- **Adaptive Layer Norm (AdaLN)** — the timestep `t` is embedded as a scalar
+  and generates per-block scale and gate parameters. This conditions every layer
+  on the current noise level.
+
+### Flow matching
+
+The model is trained with a **flow-matching** objective. Instead of predicting
+noise (as in DDPM), the model predicts a **velocity field** `v(z_t, t)` that
+defines the ODE:
+
+```
+dz/dt = v(z_t, t)
+```
+
+At inference time, we start from pure Gaussian noise `z_1` and integrate
+backward to `z_0` (the clean image) using the Euler method:
+
+```
+z_{t-dt} = z_t + v(z_t, t) * dt
+```
+
+### Noise schedule
+
+The timestep distribution follows a **logit-normal schedule** parameterized by
+`(mu, sigma)`. The mean `mu` controls how much time the sampler spends at
+different noise levels — higher `mu` shifts more steps toward higher noise
+(important for high-resolution images). The schedule auto-adjusts for
+resolution:
+
+```
+mu_adjusted = mu_base + 0.5 * log(num_pixels / base_pixels)
+```
+
+where `base_pixels = 512 * 512`.
+
+## 3. Classifier-Free Guidance (CFG)
+
+At each sampling step, two forward passes are run through the DiT:
+
+1. **Conditional (positive):** full text features + noisy image latents.
+2. **Unconditional (negative):** zeroed text features + noisy image latents
+   (image-only tokens, asymmetric CFG).
+
+The guided velocity is a weighted combination:
+
+```
+v_guided = gw * v_conditional + (1 - gw) * v_unconditional
+```
+
+where `gw` is the per-step guidance weight. With
+`gw > 1`, the model amplifies the text-conditional signal and suppresses the
+unconditional prediction, producing images that follow the prompt more
+faithfully.
+
+**Asymmetric CFG:** The unconditional branch only processes image tokens (no
+text padding), making it computationally cheaper than a full-sequence negative
+pass.
+
+**Per-step schedules:** The guidance weight can vary across steps. The
+`V4_QUALITY_48` preset, for example, uses `gw=7` for the first 45 steps and
+`gw=3` for the final 3 "polish" steps near `t=0`.
+
+
+## 4. VAE Decoder — KL Autoencoder
+
+The denoised latent `z_0` is decoded to pixel space using a frozen KL
+autoencoder.
+
+**What it does:**
+- **Unpatching:** The DiT works with 2×2 patches of latent pixels. The decoder
+  input is reshaped from `(batch, grid_h * grid_w, channels * 4)` to
+  `(batch, channels, grid_h * 2, grid_w * 2)`.
+- **Denormalization:** Per-channel shift and scale are applied to undo the
+  latent normalization used during training.
+- **Decoding:** The VAE decoder maps latents to RGB pixels.
+- **Clipping:** Output is clamped to [-1, 1] and rescaled to [0, 255] uint8.
+
+**Compression factor:** The autoencoder provides 8× spatial compression on each
+axis, and the 2×2 patching in the DiT adds another 2×. So a 1024×1024 image
+is represented as a 64×64 grid of latent tokens, each with 128 channels
+(32 base channels × 2² patch).
+
+## Putting it all together
+
+```python
+# Pseudocode for one generation call:
+
+# 1. Encode text
+text_features = qwen3_vl.encode(prompt)  # (B, L_text, D)
+
+# 2. Initialize noise
+z = torch.randn(B, grid_h * grid_w, 128)  # pure noise at t=1
+
+# 3. Euler integration from t=1 to t=0
+for step in reversed(range(num_steps)):
+    t = schedule(step)
+    s = schedule(step - 1)
+
+    # Conditional pass (text + image)
+    v_cond = dit(text_features, z, t)
+
+    # Unconditional pass (image only, zeroed text)
+    v_uncond = dit(zeros, z, t)
+
+    # CFG combination
+    v = gw[step] * v_cond + (1 - gw[step]) * v_uncond
+
+    # Euler step
+    z = z + v * (s - t)
+
+# 4. Decode to pixels
+image = vae.decode(z)
+```
diff --git a/docs/prompting.md b/docs/prompting.md
new file mode 100644
index 0000000..b44698b
--- /dev/null
+++ b/docs/prompting.md
@@ -0,0 +1,362 @@
+# Prompting Guide
+
+Ideogram 4 is trained exclusively on **structured JSON captions** (represented as string type). While the
+model can accept plain-text prompts, providing a JSON object that follows the
+caption schema gives significantly better results, especially for
+controllability, spatial layout, and style fidelity.
+
+## Plain-text vs. JSON prompts
+
+You can pass in plain-text prompts directly to the model and it will work. The
+sampling parameters come from a named preset in `ideogram4.PRESETS` (the same
+ones `run_inference.py` exposes via `--sampler-preset`), unpacked into the
+`pipe()` call:
+
+```python
+from ideogram4 import PRESETS
+
+preset = PRESETS["V4_QUALITY_48"]
+images = pipe(
+  "a golden retriever on a skateboard",
+  height=1024,
+  width=1024,
+  num_steps=preset.num_steps,
+  guidance_schedule=preset.guidance_schedule,
+  mu=preset.mu,
+  std=preset.std,
+)
+```
+
+
+But for higher quality image generations and more control, pass a JSON string as the prompt:
+
+```python
+import json
+from ideogram4 import PRESETS
+
+caption = {
+  "high_level_description": "A golden retriever riding a skateboard down a sunny sidewalk.",
+  "style_description": {
+    "aesthetics": "warm, playful, vibrant",
+    "lighting": "bright afternoon sunlight, long soft shadows",
+    "photo": "shallow depth of field, eye-level, 85mm lens",
+    "medium": "photograph",
+    "color_palette": ["#F5C542", "#87CEEB", "#4A4A4A", "#FFFFFF", "#2E8B57"]
+  },
+  "compositional_deconstruction": {
+    "background": "A sun-drenched suburban sidewalk lined with green hedges and a white picket fence. Dappled light filters through overhead trees.",
+    "elements": [
+      {"type": "obj", "bbox": [200, 300, 800, 900], "desc": "A golden retriever with a fluffy coat, standing on a red skateboard with all four paws. Its tongue is out and ears are flapping in the wind."},
+      {"type": "obj", "bbox": [250, 750, 750, 950], "desc": "A worn red skateboard with black wheels rolling along the concrete sidewalk."}
+    ]
+  }
+}
+
+preset = PRESETS["V4_QUALITY_48"]
+images = pipe(
+  json.dumps(caption, separators=(",", ":"), ensure_ascii=False),
+  height=1024,
+  width=1024,
+  num_steps=preset.num_steps,
+  guidance_schedule=preset.guidance_schedule,
+  mu=preset.mu,
+  std=preset.std,
+)
+```
+
+## Magic prompt
+
+Writing these captions by hand is optional. *Magic prompt* uses an LLM to expand
+a plain-text prompt into a full structured caption for you, so you get the
+quality of a JSON prompt from a casual one. It is enabled by default in
+`run_inference.py`; you can also call it directly:
+
+```python
+import os
+from ideogram4 import ClaudeOpusMagicPromptV1, PRESETS
+
+magic = ClaudeOpusMagicPromptV1(api_key=os.environ["MAGIC_PROMPT_API_KEY"])
+caption = magic.expand("a golden retriever on a skateboard", aspect_ratio="1:1")
+preset = PRESETS["V4_QUALITY_48"]
+images = pipe(
+  caption,
+  height=1024,
+  width=1024,
+  num_steps=preset.num_steps,
+  guidance_schedule=preset.guidance_schedule,
+  mu=preset.mu,
+  std=preset.std,
+)
+```
+
+The package ships three configurations, registered by name in
+`ideogram4.MAGIC_PROMPTS` (the keys `run_inference.py` accepts via
+`--magic-prompt-model`):
+
+| Config class | Registry key | Backend |
+| :--- | :--- | :--- |
+| `Ideogram4MagicPromptV1` | `ideogram-4-v1` | Ideogram's hosted magic-prompt API (free; reads `IDEOGRAM_API_KEY`) |
+| `ClaudeOpusMagicPromptV1` | `claude-opus-v1` | [OpenRouter](https://openrouter.ai) (reads `MAGIC_PROMPT_API_KEY`) |
+| `ClaudeSonnetMagicPromptV1` | `claude-sonnet-v1` | [OpenRouter](https://openrouter.ai) (reads `MAGIC_PROMPT_API_KEY`) |
+
+`ideogram-4-v1` is the default and is **free**. It runs the expansion
+server-side, so there is no local model or system prompt involved — it just needs
+an Ideogram API key (get one at
+[developer.ideogram.ai](https://developer.ideogram.ai)). The `claude-*`
+configurations instead send one of our open-source system prompt to an OpenRouter model;
+select one with `--magic-prompt-model` and export `MAGIC_PROMPT_API_KEY`:
+
+```bash
+python run_inference.py \
+  --prompt "an isometric illustration of a tiny city floating in the clouds" \
+  --output out.png \
+  --quantization "nf4" \
+  --magic-prompt-model claude-opus-v1 \
+  --magic-prompt-key "$MAGIC_PROMPT_API_KEY"
+```
+
+See the README's [CLI](../README.md#cli) section for the rest of the flags.
+
+Our magic-prompt system prompts are **open source** (they ship in
+`src/ideogram4/magic_prompt_system_prompts/`), so you're also welcome to
+construct the caption with any system prompt and LLM of your choosing.
+
+**A few caveats:**
+
+- At Ideogram we've tested this magic prompt with **Claude Opus**. You're welcome
+  to implement your own `MagicPrompt` configurations and/or drive a different LLM
+  with our system prompt, but those paths aren't tested by us and quality may
+  vary.
+- The magic prompt shipped here is **not** the same magic prompt used in
+  production at [Ideogram.ai](https://ideogram.ai) — results will differ from the
+  hosted product (including the `ideogram-4-v1` API).
+
+## JSON caption schema
+
+> **Note:** Following this schema is **not required** — the model accepts any
+> string as a prompt. The schema below describes the exact structure the model
+> was trained on, and matching it minimizes train/eval mismatch so the model
+> generates closer to its full quality. Treat the "required" / "must" language
+> in the rest of this section as the format the [`CaptionVerifier`](../src/ideogram4/caption_verifier.py)
+> checks against, not as a hard pipeline constraint. Deviating from the schema
+> is allowed; it just means you're sampling outside the training distribution.
+
+The full caption schema has three top-level fields:
+
+1. `high_level_description` — optional string, but strongly recommended.
+2. `style_description` — optional object.
+3. `compositional_deconstruction` — **required** object.
+
+`compositional_deconstruction` must always be present. Within it, both
+`background` and `elements` are required.
+
+### `high_level_description`
+
+A one- or two-sentence summary of the entire image. Strongly recommended in every prompt.
+
+```json
+"high_level_description": "A medium-shot photograph of a barista pouring latte art in a cozy cafe."
+```
+
+### `style_description`
+
+Controls the visual style, lighting, medium, and color palette.
+
+`style_description` must contain **exactly one** of:
+
+- `photo` — for photographic captions (paired with `medium: "photograph"`).
+- `art_style` — for non-photographic captions (illustration, painting, 3D render, etc.).
+
+`aesthetics`, `lighting`, and `medium` are also required when `style_description` is present. `color_palette` is optional.
+
+**Key order is strict** and depends on which of `photo` / `art_style` is used:
+
+| Caption type | Required key order |
+| :----------- | :----------------- |
+| Photo (uses `photo`) | `aesthetics`, `lighting`, `photo`, `medium`, `color_palette` |
+| Non-photo (uses `art_style`) | `aesthetics`, `lighting`, `medium`, `art_style`, `color_palette` |
+
+`color_palette` is the only field in this list that may be omitted; if it is included it must remain in the final position.
+
+Field descriptions:
+
+| Field | Type | Description |
+| :---- | :--- | :---------- |
+| `aesthetics` | string | Aesthetic keywords (e.g. "moody, cinematic, desaturated") |
+| `lighting` | string | Lighting description (e.g. "golden hour, rim light, dramatic shadows") |
+| `photo` | string | Camera/lens details for photographic outputs (e.g. "35mm, f/1.4, bokeh"). Use this OR `art_style`, not both. |
+| `medium` | string | Medium type: `"photograph"`, `"illustration"`, `"3d_render"`, `"painting"`, `"graphic_design"`, etc. |
+| `art_style` | string | Art style description for non-photo captions (e.g. "flat vector illustration, bold outlines"). Use this OR `photo`, not both. |
+| `color_palette` | list[str] | Hex color codes that steer the image's dominant colors. Up to 16 entries. |
+
+### `compositional_deconstruction`
+
+Provides fine-grained spatial control over the image layout using bounding
+boxes and per-element descriptions. Both fields below are required.
+
+| Field | Type | Description |
+| :---- | :--- | :---------- |
+| `background` | string | Description of the background/environment (required) |
+| `elements` | list[dict] | List of elements with optional bounding boxes (required) |
+
+`background` must come before `elements`.
+
+Each element in `elements` must follow a fixed **key order** depending on its
+type. `bbox` and `color_palette` are optional within an element; if present they
+must appear in the positions shown below.
+
+| Type | Required key order |
+| :--- | :----------------- |
+| `"obj"` | `type`, `bbox`, `desc`, `color_palette` |
+| `"text"` | `type`, `bbox`, `text`, `desc`, `color_palette` |
+
+Field descriptions:
+
+| Field | Type | Description |
+| :---- | :--- | :---------- |
+| `type` | string | `"obj"` for objects/subjects, `"text"` for in-image text |
+| `bbox` | list[int] | `[y_min, x_min, y_max, x_max]` in normalized `0–1000` coordinates (origin at top-left). Optional. |
+| `desc` | string | Detailed description of the element |
+| `text` | string | (only for `type: "text"`) The literal text to render |
+| `color_palette` | list[str] | Optional per-element palette. Up to 5 hex entries. |
+
+**Key ordering matters.** The model was trained on JSON with a consistent key
+order, so maintaining it improves generation quality. The pipeline runs
+[`CaptionVerifier`](../src/ideogram4/caption_verifier.py) on every prompt and emits
+warnings for unknown keys, missing required keys, or out-of-order keys.
+
+**Hex color format.** Colors in `color_palette` must be uppercase
+`#RRGGBB` strings (e.g. `#1B1B2F`, not `#1b1b2f` or `#fff`).
+
+**Encoding.** When serializing with Python's `json` module, pass
+`separators=(",", ":")` and `ensure_ascii=False`.
+`CaptionVerifier` warns when it detects `\uXXXX` escapes with no literal
+non-ASCII characters in the raw text.
+
+## Color palette conditioning
+
+One of Ideogram 4's distinctive features is **color palette control**. By
+providing a `color_palette` array of hex colors in `style_description`, you
+can steer the dominant colors of the generated image.
+
+```json
+"style_description": {
+  "aesthetics": "moody, cinematic",
+  "lighting": "low-key, deep shadows",
+  "photo": "35mm, f/1.4",
+  "medium": "photograph",
+  "color_palette": ["#1B1B2F", "#162447", "#1F4068", "#E43F5A", "#F5F5F5"]
+}
+```
+
+Tips for effective color palette use:
+
+- **Up to 16 colors** in `style_description.color_palette` for the overall
+  image palette, and **up to 5 colors** per element in
+  `compositional_deconstruction.elements[*].color_palette`.
+- **Include background colors** — if you want a dark background, include the
+  dark hex in the palette.
+- **Contrast pairs** — include both your highlight and shadow colors for more
+  controlled lighting.
+- **Uppercase hex only** — `#RRGGBB` form, no shorthand.
+
+### Example: warm sunset palette
+
+```json
+{
+  "high_level_description": "A lone sailboat on calm water at sunset.",
+  "style_description": {
+    "aesthetics": "serene, warm, golden hour",
+    "lighting": "golden hour backlighting, warm atmospheric haze",
+    "photo": "wide angle, f/8, long exposure",
+    "medium": "photograph",
+    "color_palette": ["#FF6B35", "#F7C59F", "#004E89", "#1A659E", "#2B2D42"]
+  },
+  "compositional_deconstruction": {
+    "background": "A calm ocean stretching to a low horizon, sky washed in orange and pink with thin wisps of cloud.",
+    "elements": [
+      {"type": "obj", "desc": "A single sailboat with a white triangular sail, silhouetted against the setting sun."}
+    ]
+  }
+}
+```
+
+
+### Example: corporate design palette
+
+```json
+{
+  "high_level_description": "A clean, modern business card layout for a tech company.",
+  "style_description": {
+    "aesthetics": "minimal, professional, geometric",
+    "lighting": "even, diffuse studio lighting",
+    "medium": "graphic_design",
+    "art_style": "flat vector design, generous whitespace, sans-serif typography",
+    "color_palette": ["#FFFFFF", "#F0F0F0", "#333333", "#0066FF", "#00CC88"]
+  },
+  "compositional_deconstruction": {
+    "background": "A solid off-white card surface with subtle paper texture.",
+    "elements": [
+      {"type": "text", "text": "ACME TECH", "desc": "Bold dark grey sans-serif company name across the upper third of the card."},
+      {"type": "text", "text": "hello@acme.tech", "desc": "Small blue sans-serif contact email near the bottom of the card."}
+    ]
+  }
+}
+```
+
+
+
+## Full example
+
+```json
+{
+  "high_level_description": "A medium-shot photograph of Formula 1 driver Max Verstappen wearing his Red Bull Racing racing suit and cap, smiling as he holds his racing helmet and talks to a man in a white shirt and black vest at a race track.",
+  "style_description": {
+    "aesthetics": "saturated primary colors, rule of thirds, joyful and triumphant",
+    "lighting": "overcast daylight, diffused, soft subtle shadows",
+    "photo": "shallow depth of field, sharp focus, eye-level, telephoto",
+    "medium": "photograph"
+  },
+  "compositional_deconstruction": {
+    "background": "The background is an out-of-focus racing paddock or track environment. Several blurred figures are visible, including one in an orange shirt. A purple and white structure with a red 'F1' logo stands on the left. The scene is outdoors with daylight, though the sky is not visible.",
+    "elements": [
+      {"type": "obj", "bbox": [55, 642, 1000, 937], "desc": "An older man standing in profile, facing left toward Max Verstappen. He has grey hair and fair skin. He is wearing a white long-sleeved button-down shirt with a navy blue quilted vest over it. He has a slight smile."},
+      {"type": "obj", "bbox": [34, 137, 1000, 617], "desc": "Max Verstappen, a fair-skinned male Formula 1 driver, positioned in the center. He is facing forward with a joyful expression and a slight smile. He wears a navy blue Red Bull Racing team uniform with numerous sponsor logos and a matching baseball cap with the number '1'. He is holding a white and red racing helmet in his hands. He has a silver watch on his left wrist."},
+      {"type": "obj", "bbox": [422, 212, 792, 452], "desc": "Max Verstappen's racing helmet, held in front of his chest. It features a white, red, and yellow design with the Red Bull logo and the 'Player 0.0' branding. The visor is clear and open."},
+      {"type": "text", "bbox": [657, 0, 755, 142], "text": "F1", "desc": "Large, stylized red logo on a black and purple background in the lower left."},
+      {"type": "text", "bbox": [768, 0, 818, 147], "text": "Formula 1\nWorld Championship™", "desc": "Small white sans-serif text below the F1 logo on the left side."},
+      {"type": "text", "bbox": [78, 447, 117, 510], "text": "ORACLE\nRed Bull\nRacing", "desc": "Very small white and orange logo on the front of the navy blue cap."},
+      {"type": "text", "bbox": [78, 417, 120, 440], "text": "1", "desc": "Bold red numeral '1' on the front left side of the navy blue cap."},
+      {"type": "text", "bbox": [332, 442, 363, 483], "text": "Red Bull", "desc": "Small yellow and red text logo on the collar of the uniform."},
+      {"type": "text", "bbox": [373, 490, 423, 532], "text": "RAUCH", "desc": "Small yellow and blue logo on the right chest of the uniform."},
+      {"type": "text", "bbox": [422, 473, 500, 532], "text": "BYBIT\nHONDA", "desc": "Medium-sized white sans-serif text on the right chest of the uniform."},
+      {"type": "text", "bbox": [410, 203, 442, 257], "text": "RAUCH", "desc": "Small yellow logo on the left upper arm of the uniform."},
+      {"type": "text", "bbox": [530, 448, 627, 510], "text": "Red Bull", "desc": "Medium red text logo on the right side of the torso, part of the Red Bull graphic."},
+      {"type": "text", "bbox": [680, 417, 768, 523], "text": "Red Bull", "desc": "Large red text logo across the lower torso of the uniform."},
+      {"type": "text", "bbox": [797, 475, 815, 518], "text": "MAX", "desc": "Small white text next to a Dutch flag on the belt area of the uniform."},
+      {"type": "text", "bbox": [558, 317, 715, 355], "text": "Player 0.0", "desc": "Black sans-serif text on a white band on the racing helmet."},
+      {"type": "text", "bbox": [560, 800, 582, 835], "text": "IA.COM", "desc": "Small blue sans-serif text on the right sleeve of the white shirt."},
+      {"type": "text", "bbox": [968, 8, 997, 332], "text": "© Anadolu Agency via Getty Images", "desc": "Small white watermark text in the bottom left corner."}
+    ]
+  }
+}
+```
+
+## Safety filter
+
+NSFW prompts are blocked. Instead of an image, the model returns a gray screen
+with the text "Image blocked by safety filter". False positive rates for safety
+is higher for non-json like prompts. We are aware that this is an issue an we may
+make a future checkpoint update to improve it.
+
+# Congratulations!
+
+You are now a certified Ideogram 4 prompter!
+
+With structured JSON captions, you have fine-grained control over composition,
+color palettes, typography, and spatial layout — capabilities that go far
+beyond what plain-text prompts can express!
+We'd love to see what you create :-)
+Share your results, experiments, and creative discoveries with the community,
+especially the unexpected ones. Tag us on social media or open a discussion on
+the repo. Happy generating!
diff --git a/eng-vlack.png b/eng-vlack.png
new file mode 100644
index 0000000..780780b
Binary files /dev/null and b/eng-vlack.png differ
diff --git a/ideogram4NSFWComfyui_v11.json b/ideogram4NSFWComfyui_v11.json
new file mode 100644
index 0000000..5965e48
--- /dev/null
+++ b/ideogram4NSFWComfyui_v11.json
@@ -0,0 +1,2853 @@
+{
+  "id": "fdcfc2b2-168f-4f3d-a1d0-6986b802fce5",
+  "revision": 0,
+  "last_node_id": 183,
+  "last_link_id": 259,
+  "nodes": [
+    {
+      "id": 99,
+      "type": "MarkdownNote",
+      "pos": [
+        4570,
+        530
+      ],
+      "size": [
+        530,
+        990
+      ],
+      "flags": {},
+      "order": 0,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [],
+      "title": "Note: Model link",
+      "properties": {
+        "ue_properties": {
+          "widget_ue_connectable": {},
+          "version": "7.8",
+          "input_ue_unconnectable": {}
+        }
+      },
+      "widgets_values": [
+        "Guide: [Subgraph](https://docs.comfy.org/interface/features/subgraph)\n\n## Model Links (for Local Users)\n\n**vae**\n\n- [flux2-vae.safetensors](https://huggingface.co/Comfy-Org/flux2-dev/resolve/main/split_files/vae/flux2-vae.safetensors)\n\n**diffusion_models**\n\n- [ideogram4_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Ideogram-4/resolve/main/diffusion_models/ideogram4_fp8_scaled.safetensors)\n- [ideogram4_unconditional_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Ideogram-4/resolve/main/diffusion_models/ideogram4_unconditional_fp8_scaled.safetensors)\n\n**text_encoders**\n\n- [qwen3vl_8b_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Qwen3-VL/resolve/main/text_encoders/qwen3vl_8b_fp8_scaled.safetensors)\n- [gemma4_e4b_it_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/gemma-4/resolve/main/text_encoders/gemma4_e4b_it_fp8_scaled.safetensors)\n\n\n## Model Storage Location\n\n```\n📂 ComfyUI/\n├── 📂 models/\n│   ├── 📂 vae/\n│   │   └── flux2-vae.safetensors\n│   ├── 📂 diffusion_models/\n│   │   ├── ideogram4_fp8_scaled.safetensors\n│   │   └── ideogram4_unconditional_fp8_scaled.safetensors\n│   └── 📂 text_encoders/\n│       ├── qwen3vl_8b_fp8_scaled.safetensors\n│       └── gemma4_e4b_it_fp8_scaled.safetensors\n```\n\n## Report Issue\n\nNote: Please update ComfyUI first ([guide](https://docs.comfy.org/installation/update_comfyui)) and prepare required models. Desktop/Cloud updates follow stable releases, so some nightly-supported models may not be available yet.\n\n- Cannot run / runtime errors: [ComfyUI/issues](https://github.com/comfyanonymous/ComfyUI/issues)\n- UI / frontend issues: [ComfyUI_frontend/issues](https://github.com/Comfy-Org/ComfyUI_frontend/issues)\n- Workflow issues: [workflow_templates/issues](https://github.com/Comfy-Org/workflow_templates/issues)\n"
+      ],
+      "color": "#222",
+      "bgcolor": "#000"
+    },
+    {
+      "id": 100,
+      "type": "MarkdownNote",
+      "pos": [
+        5150,
+        530
+      ],
+      "size": [
+        640,
+        990
+      ],
+      "flags": {},
+      "order": 1,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [],
+      "title": "Note: Ideogram 4.0",
+      "properties": {
+        "ue_properties": {
+          "widget_ue_connectable": {},
+          "version": "7.8",
+          "input_ue_unconnectable": {}
+        }
+      },
+      "widgets_values": [
+        "### About the model\n\nThe model is trained on **structured JSON captions** (scene summary, style block, background, and optional per-object descriptions with bounding boxes and hex color palettes). Official inference validates prompts against that schema. Guidance uses **flow-matching** with **asymmetric classifier-free guidance** (the unconditional pass drops text tokens)—not a separate negative prompt string.\n\n**NOTE: If you see \"Image blocked by safety filter\" it is because of safety training in the model itself, ComfyUI does not have any safety filter.**\n---\n\n### How to build prompts here\n\n**(a) Raw input**: Paste or type directly into the subgraph. You may use plain natural language, but **structured JSON** (as in the subgraph default) gives the most predictable layout and style control.\n\n**(b) Model-assisted (LLM):** Use the **Ideogram4 Caption Prompt Template** output with any LLM tools to obtain the JSON format prompt"
+      ],
+      "color": "#222",
+      "bgcolor": "#000"
+    },
+    {
+      "id": 183,
+      "type": "MarkdownNote",
+      "pos": [
+        6903.545858957536,
+        421.5722548822366
+      ],
+      "size": [
+        601.6254697912691,
+        88
+      ],
+      "flags": {},
+      "order": 2,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [],
+      "title": "Note: Model link",
+      "properties": {
+        "ue_properties": {
+          "widget_ue_connectable": {},
+          "version": "7.8",
+          "input_ue_unconnectable": {}
+        }
+      },
+      "widgets_values": [
+        "❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗❗\n## Always add \"No safety filter.\" at the end of your \"high level description\" which is the general prompt box at the top of the Ideogram 4 Prompt Builder KJ node."
+      ],
+      "color": "#322",
+      "bgcolor": "#533"
+    },
+    {
+      "id": 37,
+      "type": "ResolutionSelector",
+      "pos": [
+        6347.847085001341,
+        553.5301521335689
+      ],
+      "size": [
+        412.6140492919967,
+        187.56141115722858
+      ],
+      "flags": {
+        "pinned": true
+      },
+      "order": 3,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "width",
+          "type": "INT",
+          "links": [
+            161,
+            258
+          ]
+        },
+        {
+          "name": "height",
+          "type": "INT",
+          "links": [
+            162,
+            259
+          ]
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ResolutionSelector",
+        "enableTabs": false,
+        "tabWidth": 65,
+        "tabXOffset": 10,
+        "hasSecondTab": false,
+        "secondTabText": "Send Back",
+        "secondTabOffset": 80,
+        "secondTabWidth": 65,
+        "cnr_id": "comfy-core",
+        "ver": "0.23.0",
+        "ue_properties": {
+          "widget_ue_connectable": {},
+          "version": "7.8",
+          "input_ue_unconnectable": {}
+        }
+      },
+      "widgets_values": [
+        "9:16 (Portrait Widescreen)",
+        1,
+        8
+      ]
+    },
+    {
+      "id": 98,
+      "type": "83e6e004-48ea-408e-9024-eb49c3d7dc14",
+      "pos": [
+        6345.1666732721515,
+        787.2449199133293
+      ],
+      "size": [
+        414.15963786864995,
+        397.10074962765157
+      ],
+      "flags": {
+        "pinned": true
+      },
+      "order": 6,
+      "mode": 0,
+      "inputs": [
+        {
+          "label": "prompt",
+          "name": "text",
+          "type": "STRING",
+          "widget": {
+            "name": "text"
+          },
+          "link": 257
+        },
+        {
+          "label": "width",
+          "name": "value",
+          "type": "INT",
+          "widget": {
+            "name": "value"
+          },
+          "link": 161
+        },
+        {
+          "label": "height",
+          "name": "value_1",
+          "type": "INT",
+          "widget": {
+            "name": "value_1"
+          },
+          "link": 162
+        },
+        {
+          "label": "unconditional_unet",
+          "name": "unet_name_1",
+          "type": "COMBO",
+          "widget": {
+            "name": "unet_name_1"
+          },
+          "link": null
+        },
+        {
+          "label": "mode",
+          "name": "choice",
+          "type": "COMBO",
+          "widget": {
+            "name": "choice"
+          },
+          "link": null
+        }
+      ],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            255
+          ]
+        }
+      ],
+      "properties": {
+        "proxyWidgets": [
+          [
+            "24",
+            "text"
+          ],
+          [
+            "27",
+            "value"
+          ],
+          [
+            "28",
+            "value"
+          ],
+          [
+            "18",
+            "noise_seed"
+          ],
+          [
+            "23",
+            "unet_name"
+          ],
+          [
+            "177",
+            "clip_name"
+          ],
+          [
+            "9",
+            "vae_name"
+          ],
+          [
+            "154",
+            "unet_name"
+          ],
+          [
+            "156",
+            "choice"
+          ]
+        ],
+        "enableTabs": false,
+        "tabWidth": 65,
+        "tabXOffset": 10,
+        "hasSecondTab": false,
+        "secondTabText": "Send Back",
+        "secondTabOffset": 80,
+        "secondTabWidth": 65,
+        "cnr_id": "comfy-core",
+        "ver": "0.23.0",
+        "ue_properties": {
+          "widget_ue_connectable": {
+            "text": true,
+            "value": true,
+            "value_1": true,
+            "unet_name_1": true,
+            "choice": true
+          },
+          "version": "7.8",
+          "input_ue_unconnectable": {}
+        }
+      },
+      "widgets_values": []
+    },
+    {
+      "id": 179,
+      "type": "Ideogram4PromptBuilderKJ",
+      "pos": [
+        6777.4264015679855,
+        554.287244307294
+      ],
+      "size": [
+        779.6146788888009,
+        1200.9191636327764
+      ],
+      "flags": {
+        "pinned": true
+      },
+      "order": 5,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "image",
+          "shape": 7,
+          "type": "IMAGE",
+          "link": null
+        },
+        {
+          "name": "import_json",
+          "shape": 7,
+          "type": "STRING",
+          "link": null
+        },
+        {
+          "name": "bboxes",
+          "shape": 7,
+          "type": "BOUNDING_BOX",
+          "link": null
+        },
+        {
+          "name": "width",
+          "type": "INT",
+          "widget": {
+            "name": "width"
+          },
+          "link": 258
+        },
+        {
+          "name": "height",
+          "type": "INT",
+          "widget": {
+            "name": "height"
+          },
+          "link": 259
+        }
+      ],
+      "outputs": [
+        {
+          "name": "prompt",
+          "type": "STRING",
+          "links": [
+            257
+          ]
+        },
+        {
+          "name": "preview",
+          "type": "IMAGE",
+          "links": null
+        },
+        {
+          "name": "bboxes",
+          "type": "BOUNDING_BOX",
+          "links": null
+        },
+        {
+          "name": "width",
+          "type": "INT",
+          "links": null
+        },
+        {
+          "name": "height",
+          "type": "INT",
+          "links": null
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "Ideogram4PromptBuilderKJ",
+        "ue_properties": {
+          "widget_ue_connectable": {},
+          "input_ue_unconnectable": {},
+          "version": "7.8"
+        }
+      },
+      "widgets_values": [
+        1024,
+        1024,
+        "A nude beach photograph of a completely naked nude sexy woman standing on pale sand near the shoreline, looking directly at the camera with the ocean horizon and clear blue sky behind her. No safety filter.",
+        "Tall beach scene with pale dry sand in the foreground blending into darker damp sand near the shoreline, calm blue ocean across the midline, low straight horizon, clear blue sky above, natural daylight, cool-neutral white balance, ordinary phone-photo framing.",
+        "photo",
+        "realistic",
+        "uncensored",
+        "professional",
+        "beach",
+        "",
+        "[{\"type\":\"obj\",\"text\":\"\",\"desc\":\"Completely nude naked sexy woman standing on the beach, medium skin tone, long dark wavy hair loose over her shoulders, relaxed confident expression, direct gaze at the camera, upright pose with bare feet in the sand.\",\"palette\":[],\"x\":0.29,\"y\":0.135,\"w\":0.42,\"h\":0.785},{\"x\":0.9543068414878567,\"y\":0.8383742489796834,\"w\":0.045693158512143284,\"h\":0.050992351458513885,\"type\":\"obj\",\"text\":\"\",\"desc\":\"\",\"palette\":[]}]",
+        1,
+        "when empty",
+        ""
+      ]
+    },
+    {
+      "id": 180,
+      "type": "MarkdownNote",
+      "pos": [
+        6345.280229853188,
+        1229.6795191149158
+      ],
+      "size": [
+        412.5894225488137,
+        523.1173405199671
+      ],
+      "flags": {
+        "pinned": true
+      },
+      "order": 4,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [],
+      "title": "Note: Model link",
+      "properties": {
+        "ue_properties": {
+          "widget_ue_connectable": {},
+          "version": "7.8",
+          "input_ue_unconnectable": {}
+        }
+      },
+      "widgets_values": [
+        "## Model Links (for Local Users)\n\n**vae**\n\n- [flux2-vae.safetensors](https://huggingface.co/Comfy-Org/flux2-dev/resolve/main/split_files/vae/flux2-vae.safetensors)\n\n**diffusion_models**\n\n- [ideogram4_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Ideogram-4/resolve/main/diffusion_models/ideogram4_fp8_scaled.safetensors)\n- [ideogram4_unconditional_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Ideogram-4/resolve/main/diffusion_models/ideogram4_unconditional_fp8_scaled.safetensors)\n\n**text_encoders**\n\n- [Qwen3VL-8B-Uncensored-HauhauCS-Aggressive](https://huggingface.co/HauhauCS/Qwen3VL-8B-Uncensored-HauhauCS-Aggressive) / Pick Q4 for low VRAM\n- [gemma4_e4b_it_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/gemma-4/resolve/main/text_encoders/gemma4_e4b_it_fp8_scaled.safetensors)\n\n\n## Model Storage Location\n\n```\n📂 ComfyUI/\n├── 📂 models/\n│   ├── 📂 vae/\n│   │   └── flux2-vae.safetensors\n│   ├── 📂 diffusion_models/\n│   │   ├── ideogram4_fp8_scaled.safetensors\n│   │   └── ideogram4_unconditional_fp8_scaled.safetensors\n│   └── 📂 text_encoders/\n│       ├── Qwen3VL-8B-Uncensored-HauhauCS-Aggressive-Q*.gguf\n│       └── gemma4_e4b_it_fp8_scaled.safetensors\n```\n\n## Report Issue\n\nNote: Please update ComfyUI first ([guide](https://docs.comfy.org/installation/update_comfyui)) and prepare required models. Desktop/Cloud updates follow stable releases, so some nightly-supported models may not be available yet.\n\n- Cannot run / runtime errors: [ComfyUI/issues](https://github.com/comfyanonymous/ComfyUI/issues)\n- UI / frontend issues: [ComfyUI_frontend/issues](https://github.com/Comfy-Org/ComfyUI_frontend/issues)\n- Workflow issues: [workflow_templates/issues](https://github.com/Comfy-Org/workflow_templates/issues)\n"
+      ],
+      "color": "#222",
+      "bgcolor": "#000"
+    },
+    {
+      "id": 178,
+      "type": "PreviewImage",
+      "pos": [
+        7575.996537923663,
+        554.8233299178336
+      ],
+      "size": [
+        820.6441510853892,
+        1199.1879392761516
+      ],
+      "flags": {
+        "pinned": true
+      },
+      "order": 7,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "link": 255
+        }
+      ],
+      "outputs": [],
+      "properties": {
+        "Node name for S&R": "PreviewImage",
+        "ue_properties": {
+          "widget_ue_connectable": {},
+          "input_ue_unconnectable": {},
+          "version": "7.8"
+        }
+      },
+      "widgets_values": []
+    }
+  ],
+  "links": [
+    [
+      161,
+      37,
+      0,
+      98,
+      1,
+      "INT"
+    ],
+    [
+      162,
+      37,
+      1,
+      98,
+      2,
+      "INT"
+    ],
+    [
+      255,
+      98,
+      0,
+      178,
+      0,
+      "IMAGE"
+    ],
+    [
+      257,
+      179,
+      0,
+      98,
+      0,
+      "STRING"
+    ],
+    [
+      258,
+      37,
+      0,
+      179,
+      3,
+      "INT"
+    ],
+    [
+      259,
+      37,
+      1,
+      179,
+      4,
+      "INT"
+    ]
+  ],
+  "groups": [],
+  "definitions": {
+    "subgraphs": [
+      {
+        "id": "83e6e004-48ea-408e-9024-eb49c3d7dc14",
+        "version": 1,
+        "state": {
+          "lastGroupId": 9,
+          "lastNodeId": 183,
+          "lastLinkId": 259,
+          "lastRerouteId": 0
+        },
+        "revision": 0,
+        "config": {},
+        "name": "Text to Image (Ideogram v4)",
+        "inputNode": {
+          "id": -10,
+          "bounding": [
+            3490,
+            920,
+            154.9000015258789,
+            228
+          ]
+        },
+        "outputNode": {
+          "id": -20,
+          "bounding": [
+            6850,
+            936,
+            128,
+            68
+          ]
+        },
+        "inputs": [
+          {
+            "id": "4bc742d1-7b4b-452c-90d9-0d76ebcdae76",
+            "name": "text",
+            "type": "STRING",
+            "linkIds": [
+              152
+            ],
+            "label": "prompt",
+            "pos": [
+              3620.900001525879,
+              944
+            ]
+          },
+          {
+            "id": "8d4038eb-73c7-45e9-bba1-f068f55e8d32",
+            "name": "value",
+            "type": "INT",
+            "linkIds": [
+              153
+            ],
+            "label": "width",
+            "pos": [
+              3620.900001525879,
+              964
+            ]
+          },
+          {
+            "id": "281550e6-6acf-4cbe-aec1-9eb803b4dec1",
+            "name": "value_1",
+            "type": "INT",
+            "linkIds": [
+              154
+            ],
+            "label": "height",
+            "pos": [
+              3620.900001525879,
+              984
+            ]
+          },
+          {
+            "id": "fae56884-2f1a-470b-a25f-40e7a87ef69d",
+            "name": "noise_seed",
+            "type": "INT",
+            "linkIds": [
+              155
+            ],
+            "pos": [
+              3620.900001525879,
+              1004
+            ]
+          },
+          {
+            "id": "3497309c-a7d7-4e28-9330-142c15881632",
+            "name": "unet_name",
+            "type": "COMBO",
+            "linkIds": [
+              156
+            ],
+            "pos": [
+              3620.900001525879,
+              1024
+            ]
+          },
+          {
+            "id": "e87126db-7147-465e-b129-370ed2c6cc22",
+            "name": "clip_name",
+            "type": "COMBO",
+            "linkIds": [
+              253
+            ],
+            "pos": [
+              3620.900001525879,
+              1044
+            ]
+          },
+          {
+            "id": "a1e6c080-b11b-4d5c-a3a8-fcf4df654cf7",
+            "name": "vae_name",
+            "type": "COMBO",
+            "linkIds": [
+              158
+            ],
+            "pos": [
+              3620.900001525879,
+              1064
+            ]
+          },
+          {
+            "id": "b0d16516-95de-44d9-bea8-3cd2e7c78e9a",
+            "name": "unet_name_1",
+            "type": "COMBO",
+            "linkIds": [
+              216
+            ],
+            "label": "unconditional_unet",
+            "pos": [
+              3620.900001525879,
+              1084
+            ]
+          },
+          {
+            "id": "249fd825-e6b3-489d-a341-6d8050500f5e",
+            "name": "choice",
+            "type": "COMBO",
+            "linkIds": [
+              219
+            ],
+            "label": "mode",
+            "pos": [
+              3620.900001525879,
+              1104
+            ]
+          }
+        ],
+        "outputs": [
+          {
+            "id": "b81e4f60-e543-4f02-875b-b0f1bdc274f2",
+            "name": "IMAGE",
+            "type": "IMAGE",
+            "linkIds": [
+              25
+            ],
+            "localized_name": "IMAGE",
+            "pos": [
+              6874,
+              960
+            ]
+          }
+        ],
+        "widgets": [],
+        "nodes": [
+          {
+            "id": 9,
+            "type": "VAELoader",
+            "pos": [
+              4730,
+              1220
+            ],
+            "size": [
+              470,
+              110
+            ],
+            "flags": {},
+            "order": 2,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "vae_name",
+                "name": "vae_name",
+                "type": "COMBO",
+                "widget": {
+                  "name": "vae_name"
+                },
+                "link": 158
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "VAE",
+                "name": "VAE",
+                "type": "VAE",
+                "links": [
+                  17
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "VAELoader",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.8.2",
+              "models": [
+                {
+                  "name": "flux2-vae.safetensors",
+                  "url": "https://huggingface.co/Comfy-Org/flux2-dev/resolve/main/split_files/vae/flux2-vae.safetensors",
+                  "directory": "vae"
+                }
+              ],
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "flux2-vae.safetensors"
+            ]
+          },
+          {
+            "id": 10,
+            "type": "ConditioningZeroOut",
+            "pos": [
+              5450,
+              960
+            ],
+            "size": [
+              250,
+              80
+            ],
+            "flags": {
+              "collapsed": false
+            },
+            "order": 3,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "conditioning",
+                "name": "conditioning",
+                "type": "CONDITIONING",
+                "link": 8
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "CONDITIONING",
+                "name": "CONDITIONING",
+                "type": "CONDITIONING",
+                "links": [
+                  214
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "ConditioningZeroOut",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.9.1",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": []
+          },
+          {
+            "id": 11,
+            "type": "EmptyFlux2LatentImage",
+            "pos": [
+              5330,
+              1180
+            ],
+            "size": [
+              270,
+              170
+            ],
+            "flags": {},
+            "order": 4,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "width",
+                "name": "width",
+                "type": "INT",
+                "widget": {
+                  "name": "width"
+                },
+                "link": 33
+              },
+              {
+                "localized_name": "height",
+                "name": "height",
+                "type": "INT",
+                "widget": {
+                  "name": "height"
+                },
+                "link": 36
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "LATENT",
+                "name": "LATENT",
+                "type": "LATENT",
+                "links": [
+                  15
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "EmptyFlux2LatentImage",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.8.2",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              1024,
+              1024,
+              1
+            ]
+          },
+          {
+            "id": 12,
+            "type": "SamplerCustomAdvanced",
+            "pos": [
+              6160,
+              500
+            ],
+            "size": [
+              290,
+              170
+            ],
+            "flags": {},
+            "order": 5,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "noise",
+                "name": "noise",
+                "type": "NOISE",
+                "link": 11
+              },
+              {
+                "localized_name": "guider",
+                "name": "guider",
+                "type": "GUIDER",
+                "link": 215
+              },
+              {
+                "localized_name": "sampler",
+                "name": "sampler",
+                "type": "SAMPLER",
+                "link": 13
+              },
+              {
+                "localized_name": "sigmas",
+                "name": "sigmas",
+                "type": "SIGMAS",
+                "link": 14
+              },
+              {
+                "localized_name": "latent_image",
+                "name": "latent_image",
+                "type": "LATENT",
+                "link": 15
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "output",
+                "name": "output",
+                "type": "LATENT",
+                "links": [
+                  16
+                ]
+              },
+              {
+                "localized_name": "denoised_output",
+                "name": "denoised_output",
+                "type": "LATENT",
+                "links": []
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "SamplerCustomAdvanced",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.8.2",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": []
+          },
+          {
+            "id": 13,
+            "type": "VAEDecode",
+            "pos": [
+              6560,
+              500
+            ],
+            "size": [
+              230,
+              100
+            ],
+            "flags": {},
+            "order": 6,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "samples",
+                "name": "samples",
+                "type": "LATENT",
+                "link": 16
+              },
+              {
+                "localized_name": "vae",
+                "name": "vae",
+                "type": "VAE",
+                "link": 17
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "IMAGE",
+                "name": "IMAGE",
+                "type": "IMAGE",
+                "slot_index": 0,
+                "links": [
+                  25
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "VAEDecode",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.8.2",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": []
+          },
+          {
+            "id": 16,
+            "type": "KSamplerSelect",
+            "pos": [
+              5790,
+              1100
+            ],
+            "size": [
+              270,
+              110
+            ],
+            "flags": {},
+            "order": 0,
+            "mode": 0,
+            "inputs": [],
+            "outputs": [
+              {
+                "localized_name": "SAMPLER",
+                "name": "SAMPLER",
+                "type": "SAMPLER",
+                "links": [
+                  13
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "KSamplerSelect",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.8.2",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "euler"
+            ]
+          },
+          {
+            "id": 17,
+            "type": "Ideogram4Scheduler",
+            "pos": [
+              5790,
+              1260
+            ],
+            "size": [
+              270,
+              240
+            ],
+            "flags": {},
+            "order": 7,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "steps",
+                "name": "steps",
+                "type": "INT",
+                "widget": {
+                  "name": "steps"
+                },
+                "link": 207
+              },
+              {
+                "localized_name": "width",
+                "name": "width",
+                "type": "INT",
+                "widget": {
+                  "name": "width"
+                },
+                "link": 34
+              },
+              {
+                "localized_name": "height",
+                "name": "height",
+                "type": "INT",
+                "widget": {
+                  "name": "height"
+                },
+                "link": 37
+              },
+              {
+                "localized_name": "mu",
+                "name": "mu",
+                "type": "FLOAT",
+                "widget": {
+                  "name": "mu"
+                },
+                "link": 208
+              },
+              {
+                "localized_name": "std",
+                "name": "std",
+                "type": "FLOAT",
+                "widget": {
+                  "name": "std"
+                },
+                "link": 209
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "SIGMAS",
+                "name": "SIGMAS",
+                "type": "SIGMAS",
+                "links": [
+                  14
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "Ideogram4Scheduler",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              20,
+              1024,
+              1024,
+              0.5,
+              1.75
+            ]
+          },
+          {
+            "id": 18,
+            "type": "RandomNoise",
+            "pos": [
+              5780,
+              490
+            ],
+            "size": [
+              270,
+              110
+            ],
+            "flags": {},
+            "order": 8,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "noise_seed",
+                "name": "noise_seed",
+                "type": "INT",
+                "widget": {
+                  "name": "noise_seed"
+                },
+                "link": 155
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "NOISE",
+                "name": "NOISE",
+                "type": "NOISE",
+                "links": [
+                  11
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "RandomNoise",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.8.2",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              696249430419219,
+              "randomize"
+            ]
+          },
+          {
+            "id": 23,
+            "type": "UNETLoader",
+            "pos": [
+              4720,
+              520
+            ],
+            "size": [
+              470,
+              170
+            ],
+            "flags": {},
+            "order": 9,
+            "mode": 0,
+            "showAdvanced": true,
+            "inputs": [
+              {
+                "localized_name": "unet_name",
+                "name": "unet_name",
+                "type": "COMBO",
+                "widget": {
+                  "name": "unet_name"
+                },
+                "link": 156
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "MODEL",
+                "name": "MODEL",
+                "type": "MODEL",
+                "links": [
+                  222
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "UNETLoader",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.8.2",
+              "models": [
+                {
+                  "name": "ideogram4_fp8_scaled.safetensors",
+                  "url": "https://huggingface.co/Comfy-Org/Ideogram-4/resolve/main/diffusion_models/ideogram4_fp8_scaled.safetensors",
+                  "directory": "diffusion_models"
+                }
+              ],
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "ideogram4_fp8_scaled.safetensors",
+              "default"
+            ]
+          },
+          {
+            "id": 24,
+            "type": "CLIPTextEncode",
+            "pos": [
+              5270,
+              500
+            ],
+            "size": [
+              430,
+              420
+            ],
+            "flags": {},
+            "order": 10,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "clip",
+                "name": "clip",
+                "type": "CLIP",
+                "link": 254
+              },
+              {
+                "localized_name": "text",
+                "name": "text",
+                "type": "STRING",
+                "widget": {
+                  "name": "text"
+                },
+                "link": 152
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "CONDITIONING",
+                "name": "CONDITIONING",
+                "type": "CONDITIONING",
+                "slot_index": 0,
+                "links": [
+                  8,
+                  213
+                ]
+              }
+            ],
+            "title": "CLIP Text Encode (Positive Prompt)",
+            "properties": {
+              "Node name for S&R": "CLIPTextEncode",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.8.2",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "{\n    \"high_level_description\": \"A surreal streetwear mixed-media collage poster featuring a relaxed skateboarder mid-air against a vibrant blue sky, backed by giant puffy 3D letters spelling 'COMFY'. The composition blends retro magazine cutout aesthetics with grunge elements like torn paper banners and distressed red stamps, conveying an effortless, cozy vibe.\",\n    \"style_description\": {\n        \"aesthetics\": \"Retro magazine cutout style, mixed-media digital collage, high-contrast streetwear graphic, featuring rough ripped paper edges and distressed grunge textures.\",\n        \"lighting\": \"High-contrast flash mixed with harsh midday sunlight on the skater cutout, contrasting with flat, bright graphic lighting on the 3D typography.\",\n        \"photo\": \"Vintage grainy 35mm film with distressed halftone scan textures and subtle light leaks.\",\n        \"medium\": \"Mixed-media digital collage\",\n        \"color_palette\": [\"#1E73BE\", \"#FDFDFD\", \"#C82A2A\", \"#657C9C\", \"#EFEFEF\"]\n    },\n    \"compositional_deconstruction\": {\n        \"background\": \"A vibrant, clear blue sky layered with a vintage grainy film texture and subtle halftone dot patterns, transitioning down to an implied pale gray concrete ramp at the very bottom edge.\",\n        \"elements\": [\n            {\n                \"type\": \"obj\",\n                \"bbox\": [128, 149, 354, 810],\n                \"desc\": \"Massive 3D puffy, inflatable white typography spelling 'COMFY'. The letters stretch across the upper half of the canvas, acting as a surreal, soft cloud-like backdrop.\",\n                \"color_palette\": [\"#FDFDFD\", \"#E0E0E0\", \"#D3DBE2\"]\n            },\n            {\n                \"type\": \"obj\",\n                \"bbox\": [459, 37, 727, 264],\n                \"desc\": \"A cluster of oversized, distressed red stamped circles and dots, applied loosely to the midground like a grunge ink stamp, partially obscuring the bottom left of the text.\",\n                \"color_palette\": [\"#C82A2A\", \"#A11D1D\"]\n            },\n            {\n                \"type\": \"obj\",\n                \"bbox\": [23, 366, 153, 666],\n                \"desc\": \"A vertically oriented, torn paper side banner pinned to the left edge. The rough-edged paper displays the bold, stamped text 'STAY COZY' in high-contrast black ink.\",\n                \"color_palette\": [\"#EFEFEF\", \"#1A1A1A\", \"#C82A2A\"]\n            },\n            {\n                \"type\": \"obj\",\n                \"bbox\": [287, 210, 756, 819],\n                \"desc\": \"A sharp photographic cutout of a skateboarder mid-air in a relaxed pose. He wears loose-fitting washed denim jeans and a plain white tee, appearing to effortlessly float above the concrete ramp. A distinct white cutout border surrounds his silhouette.\",\n                \"color_palette\": [\"#FDFDFD\", \"#657C9C\", \"#2B2B2B\", \"#DCA57D\"]\n            },\n            {\n                \"type\": \"obj\",\n                \"bbox\": [773, 39, 973, 187],\n                \"desc\": \"A surreal, miniature floating skateboard cutout, positioned playfully in the upper right sky as if defying gravity.\",\n                \"color_palette\": [\"#D2A679\", \"#2B2B2B\", \"#C82A2A\"]\n            },\n            {\n                \"type\": \"obj\",\n                \"bbox\": [105, 830, 905, 980],\n                \"desc\": \"A wide, horizontal strip of heavily textured torn paper spanning the lower third of the composition. It features the bold typographic phrase 'BEYOND THE COMFORT ZONE' intermixed with 'EFFORTLESS RIDE' alongside ripped edges that reveal the background.\",\n                \"color_palette\": [\"#EFEFEF\", \"#1A1A1A\", \"#999999\"]\n            }\n        ]\n    }\n}"
+            ]
+          },
+          {
+            "id": 27,
+            "type": "PrimitiveInt",
+            "pos": [
+              4240,
+              1610
+            ],
+            "size": [
+              270,
+              110
+            ],
+            "flags": {},
+            "order": 11,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "value",
+                "name": "value",
+                "type": "INT",
+                "widget": {
+                  "name": "value"
+                },
+                "link": 153
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "INT",
+                "name": "INT",
+                "type": "INT",
+                "links": [
+                  32
+                ]
+              }
+            ],
+            "title": "Int (Width)",
+            "properties": {
+              "Node name for S&R": "PrimitiveInt",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              1024,
+              "fixed"
+            ]
+          },
+          {
+            "id": 28,
+            "type": "PrimitiveInt",
+            "pos": [
+              4250,
+              1800
+            ],
+            "size": [
+              270,
+              110
+            ],
+            "flags": {},
+            "order": 12,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "value",
+                "name": "value",
+                "type": "INT",
+                "widget": {
+                  "name": "value"
+                },
+                "link": 154
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "INT",
+                "name": "INT",
+                "type": "INT",
+                "links": [
+                  35
+                ]
+              }
+            ],
+            "title": "Int (Height)",
+            "properties": {
+              "Node name for S&R": "PrimitiveInt",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              1024,
+              "fixed"
+            ]
+          },
+          {
+            "id": 31,
+            "type": "ComfyMathExpression",
+            "pos": [
+              5340,
+              1400
+            ],
+            "size": [
+              230,
+              80
+            ],
+            "flags": {
+              "collapsed": true
+            },
+            "order": 13,
+            "mode": 0,
+            "inputs": [
+              {
+                "label": "a",
+                "localized_name": "values.a",
+                "name": "values.a",
+                "type": "FLOAT,INT,BOOLEAN",
+                "link": 32
+              },
+              {
+                "label": "b",
+                "localized_name": "values.b",
+                "name": "values.b",
+                "shape": 7,
+                "type": "FLOAT,INT,BOOLEAN",
+                "link": null
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "FLOAT",
+                "name": "FLOAT",
+                "type": "FLOAT",
+                "links": null
+              },
+              {
+                "localized_name": "INT",
+                "name": "INT",
+                "type": "INT",
+                "links": [
+                  33,
+                  34
+                ]
+              },
+              {
+                "localized_name": "BOOL",
+                "name": "BOOL",
+                "type": "BOOLEAN",
+                "links": null
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "ComfyMathExpression",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "max(((a + 15) // 16) * 16, 256)"
+            ]
+          },
+          {
+            "id": 32,
+            "type": "ComfyMathExpression",
+            "pos": [
+              5350,
+              1470
+            ],
+            "size": [
+              230,
+              80
+            ],
+            "flags": {
+              "collapsed": true
+            },
+            "order": 14,
+            "mode": 0,
+            "inputs": [
+              {
+                "label": "a",
+                "localized_name": "values.a",
+                "name": "values.a",
+                "type": "FLOAT,INT,BOOLEAN",
+                "link": 35
+              },
+              {
+                "label": "b",
+                "localized_name": "values.b",
+                "name": "values.b",
+                "shape": 7,
+                "type": "FLOAT,INT,BOOLEAN",
+                "link": null
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "FLOAT",
+                "name": "FLOAT",
+                "type": "FLOAT",
+                "links": null
+              },
+              {
+                "localized_name": "INT",
+                "name": "INT",
+                "type": "INT",
+                "links": [
+                  36,
+                  37
+                ]
+              },
+              {
+                "localized_name": "BOOL",
+                "name": "BOOL",
+                "type": "BOOLEAN",
+                "links": null
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "ComfyMathExpression",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "max(((a + 15) // 16) * 16, 256)"
+            ]
+          },
+          {
+            "id": 144,
+            "type": "ComfyNumberConvert",
+            "pos": [
+              5920,
+              1870
+            ],
+            "size": [
+              230,
+              100
+            ],
+            "flags": {},
+            "order": 15,
+            "mode": 0,
+            "inputs": [
+              {
+                "label": "value",
+                "localized_name": "value",
+                "name": "value",
+                "type": "INT,FLOAT,STRING,BOOLEAN",
+                "link": 195
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "FLOAT",
+                "name": "FLOAT",
+                "type": "FLOAT",
+                "links": [
+                  208
+                ]
+              },
+              {
+                "localized_name": "INT",
+                "name": "INT",
+                "type": "INT",
+                "links": null
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "ComfyNumberConvert",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": []
+          },
+          {
+            "id": 145,
+            "type": "JsonExtractString",
+            "pos": [
+              5450,
+              1870
+            ],
+            "size": [
+              400,
+              200
+            ],
+            "flags": {},
+            "order": 16,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "json_string",
+                "name": "json_string",
+                "type": "STRING",
+                "widget": {
+                  "name": "json_string"
+                },
+                "link": 196
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "STRING",
+                "name": "STRING",
+                "type": "STRING",
+                "links": [
+                  195
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "JsonExtractString",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "",
+              "mu"
+            ]
+          },
+          {
+            "id": 146,
+            "type": "ComfyNumberConvert",
+            "pos": [
+              5930,
+              2110
+            ],
+            "size": [
+              230,
+              100
+            ],
+            "flags": {},
+            "order": 17,
+            "mode": 0,
+            "inputs": [
+              {
+                "label": "value",
+                "localized_name": "value",
+                "name": "value",
+                "type": "INT,FLOAT,STRING,BOOLEAN",
+                "link": 197
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "FLOAT",
+                "name": "FLOAT",
+                "type": "FLOAT",
+                "links": [
+                  209
+                ]
+              },
+              {
+                "localized_name": "INT",
+                "name": "INT",
+                "type": "INT",
+                "links": null
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "ComfyNumberConvert",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": []
+          },
+          {
+            "id": 147,
+            "type": "JsonExtractString",
+            "pos": [
+              5010,
+              1630
+            ],
+            "size": [
+              410,
+              470
+            ],
+            "flags": {},
+            "order": 18,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "key",
+                "name": "key",
+                "type": "STRING",
+                "widget": {
+                  "name": "key"
+                },
+                "link": 218
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "STRING",
+                "name": "STRING",
+                "type": "STRING",
+                "links": [
+                  199
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "JsonExtractString",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "{\n  \"Quality\": {\n    \"num_steps\": 48,\n    \"mu\": 0.0,\n    \"std\": 1.5,\n    \"preset_id\": \"V4_QUALITY_48\"\n  },\n  \"Default\": {\n    \"num_steps\": 20,\n    \"mu\": 0.0,\n    \"std\": 1.75,\n    \"preset_id\": \"V4_DEFAULT_20\"\n  },\n  \"Turbo\": {\n    \"num_steps\": 12,\n    \"mu\": 0.5,\n    \"std\": 1.75,\n    \"preset_id\": \"V4_TURBO_12\"\n  }\n}",
+              "Quality"
+            ]
+          },
+          {
+            "id": 148,
+            "type": "StringReplace",
+            "pos": [
+              5050,
+              2150
+            ],
+            "size": [
+              230,
+              40
+            ],
+            "flags": {
+              "collapsed": true
+            },
+            "order": 19,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "string",
+                "name": "string",
+                "type": "STRING",
+                "widget": {
+                  "name": "string"
+                },
+                "link": 199
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "STRING",
+                "name": "STRING",
+                "type": "STRING",
+                "links": [
+                  196,
+                  200,
+                  201
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "StringReplace",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "",
+              "'",
+              "\""
+            ]
+          },
+          {
+            "id": 149,
+            "type": "JsonExtractString",
+            "pos": [
+              5460,
+              1610
+            ],
+            "size": [
+              400,
+              200
+            ],
+            "flags": {},
+            "order": 20,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "json_string",
+                "name": "json_string",
+                "type": "STRING",
+                "widget": {
+                  "name": "json_string"
+                },
+                "link": 200
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "STRING",
+                "name": "STRING",
+                "type": "STRING",
+                "links": [
+                  202
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "JsonExtractString",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "",
+              "num_steps"
+            ]
+          },
+          {
+            "id": 150,
+            "type": "JsonExtractString",
+            "pos": [
+              5450,
+              2110
+            ],
+            "size": [
+              400,
+              200
+            ],
+            "flags": {},
+            "order": 21,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "json_string",
+                "name": "json_string",
+                "type": "STRING",
+                "widget": {
+                  "name": "json_string"
+                },
+                "link": 201
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "STRING",
+                "name": "STRING",
+                "type": "STRING",
+                "links": [
+                  197
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "JsonExtractString",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "",
+              "std"
+            ]
+          },
+          {
+            "id": 151,
+            "type": "ComfyNumberConvert",
+            "pos": [
+              5920,
+              1620
+            ],
+            "size": [
+              230,
+              100
+            ],
+            "flags": {},
+            "order": 22,
+            "mode": 0,
+            "inputs": [
+              {
+                "label": "value",
+                "localized_name": "value",
+                "name": "value",
+                "type": "INT,FLOAT,STRING,BOOLEAN",
+                "link": 202
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "FLOAT",
+                "name": "FLOAT",
+                "type": "FLOAT",
+                "links": []
+              },
+              {
+                "localized_name": "INT",
+                "name": "INT",
+                "type": "INT",
+                "links": [
+                  207
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "ComfyNumberConvert",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": []
+          },
+          {
+            "id": 154,
+            "type": "UNETLoader",
+            "pos": [
+              4730,
+              740
+            ],
+            "size": [
+              470,
+              170
+            ],
+            "flags": {},
+            "order": 23,
+            "mode": 0,
+            "showAdvanced": true,
+            "inputs": [
+              {
+                "localized_name": "unet_name",
+                "name": "unet_name",
+                "type": "COMBO",
+                "widget": {
+                  "name": "unet_name"
+                },
+                "link": 216
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "MODEL",
+                "name": "MODEL",
+                "type": "MODEL",
+                "links": [
+                  211
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "UNETLoader",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.8.2",
+              "models": [
+                {
+                  "name": "ideogram4_unconditional_fp8_scaled.safetensors",
+                  "url": "https://huggingface.co/Comfy-Org/Ideogram-4/resolve/main/diffusion_models/ideogram4_unconditional_fp8_scaled.safetensors",
+                  "directory": "diffusion_models"
+                }
+              ],
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "ideogram4_unconditional_fp8_scaled.safetensors",
+              "default"
+            ]
+          },
+          {
+            "id": 155,
+            "type": "DualModelGuider",
+            "pos": [
+              5790,
+              870
+            ],
+            "size": [
+              270,
+              180
+            ],
+            "flags": {},
+            "order": 24,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "model",
+                "name": "model",
+                "type": "MODEL",
+                "link": 223
+              },
+              {
+                "localized_name": "positive",
+                "name": "positive",
+                "type": "CONDITIONING",
+                "link": 213
+              },
+              {
+                "localized_name": "model_negative",
+                "name": "model_negative",
+                "shape": 7,
+                "type": "MODEL",
+                "link": 211
+              },
+              {
+                "localized_name": "negative",
+                "name": "negative",
+                "shape": 7,
+                "type": "CONDITIONING",
+                "link": 214
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "GUIDER",
+                "name": "GUIDER",
+                "type": "GUIDER",
+                "links": [
+                  215
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "DualModelGuider",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              7
+            ]
+          },
+          {
+            "id": 156,
+            "type": "CustomCombo",
+            "pos": [
+              4720,
+              1630
+            ],
+            "size": [
+              270,
+              280
+            ],
+            "flags": {},
+            "order": 25,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "choice",
+                "name": "choice",
+                "type": "COMBO",
+                "widget": {
+                  "name": "choice"
+                },
+                "link": 219
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "STRING",
+                "name": "STRING",
+                "type": "STRING",
+                "links": [
+                  218
+                ]
+              },
+              {
+                "localized_name": "INDEX",
+                "name": "INDEX",
+                "type": "INT",
+                "links": null
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "CustomCombo",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "Quality",
+              0,
+              "Quality",
+              "Default",
+              "Turbo",
+              ""
+            ]
+          },
+          {
+            "id": 157,
+            "type": "CFGOverride",
+            "pos": [
+              5790,
+              650
+            ],
+            "size": [
+              260,
+              170
+            ],
+            "flags": {},
+            "order": 26,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "model",
+                "name": "model",
+                "type": "MODEL",
+                "link": 222
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "MODEL",
+                "name": "MODEL",
+                "type": "MODEL",
+                "links": [
+                  223
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "CFGOverride",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.23.0",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              3,
+              0.7,
+              1
+            ]
+          },
+          {
+            "id": 14,
+            "type": "CLIPLoader",
+            "pos": [
+              4180.193534766597,
+              1037.8893403553386
+            ],
+            "size": [
+              470,
+              170
+            ],
+            "flags": {},
+            "order": 1,
+            "mode": 4,
+            "inputs": [],
+            "outputs": [
+              {
+                "localized_name": "CLIP",
+                "name": "CLIP",
+                "type": "CLIP",
+                "links": []
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "CLIPLoader",
+              "enableTabs": false,
+              "tabWidth": 65,
+              "tabXOffset": 10,
+              "hasSecondTab": false,
+              "secondTabText": "Send Back",
+              "secondTabOffset": 80,
+              "secondTabWidth": 65,
+              "cnr_id": "comfy-core",
+              "ver": "0.8.2",
+              "models": [
+                {
+                  "name": "qwen3vl_8b_fp8_scaled.safetensors",
+                  "url": "https://huggingface.co/Comfy-Org/Qwen3-VL/resolve/main/text_encoders/qwen3vl_8b_fp8_scaled.safetensors",
+                  "directory": "text_encoders"
+                }
+              ],
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "version": "7.8",
+                "input_ue_unconnectable": {}
+              }
+            },
+            "widgets_values": [
+              "qwen3vl_8b_fp8_scaled.safetensors",
+              "ideogram4",
+              "default"
+            ]
+          },
+          {
+            "id": 177,
+            "type": "CLIPLoaderGGUF",
+            "pos": [
+              4734.296665320757,
+              966.5983760744808
+            ],
+            "size": [
+              461.55736142135447,
+              189.75101579951183
+            ],
+            "flags": {},
+            "order": 27,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "clip_name",
+                "name": "clip_name",
+                "type": "COMBO",
+                "widget": {
+                  "name": "clip_name"
+                },
+                "link": 253
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "CLIP",
+                "name": "CLIP",
+                "type": "CLIP",
+                "links": [
+                  254
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "CLIPLoaderGGUF",
+              "ue_properties": {
+                "widget_ue_connectable": {},
+                "input_ue_unconnectable": {},
+                "version": "7.8"
+              }
+            },
+            "widgets_values": [
+              "Qwen3VL-8B-Uncensored-HauhauCS-Aggressive-Q8_0.gguf",
+              "ideogram4"
+            ]
+          }
+        ],
+        "groups": [
+          {
+            "id": 1,
+            "title": "Models",
+            "bounding": [
+              4700,
+              420,
+              530,
+              1100
+            ],
+            "color": "#3f789e",
+            "flags": {}
+          },
+          {
+            "id": 2,
+            "title": "Latent Size",
+            "bounding": [
+              5260,
+              1080,
+              450,
+              440
+            ],
+            "color": "#3f789e",
+            "flags": {}
+          },
+          {
+            "id": 3,
+            "title": "Sampling",
+            "bounding": [
+              5740,
+              420,
+              780,
+              1100
+            ],
+            "color": "#3f789e",
+            "flags": {}
+          },
+          {
+            "id": 4,
+            "title": "Prompt",
+            "bounding": [
+              5260,
+              420,
+              450,
+              640
+            ],
+            "color": "#3f789e",
+            "flags": {}
+          },
+          {
+            "id": 5,
+            "title": "Image Size",
+            "bounding": [
+              4130,
+              1540,
+              530,
+              420
+            ],
+            "color": "#3f789e",
+            "flags": {}
+          },
+          {
+            "id": 9,
+            "title": "Preset",
+            "bounding": [
+              4700,
+              1540,
+              1820,
+              780
+            ],
+            "color": "#3f789e",
+            "flags": {}
+          }
+        ],
+        "links": [
+          {
+            "id": 8,
+            "origin_id": 24,
+            "origin_slot": 0,
+            "target_id": 10,
+            "target_slot": 0,
+            "type": "CONDITIONING"
+          },
+          {
+            "id": 33,
+            "origin_id": 31,
+            "origin_slot": 1,
+            "target_id": 11,
+            "target_slot": 0,
+            "type": "INT"
+          },
+          {
+            "id": 36,
+            "origin_id": 32,
+            "origin_slot": 1,
+            "target_id": 11,
+            "target_slot": 1,
+            "type": "INT"
+          },
+          {
+            "id": 11,
+            "origin_id": 18,
+            "origin_slot": 0,
+            "target_id": 12,
+            "target_slot": 0,
+            "type": "NOISE"
+          },
+          {
+            "id": 13,
+            "origin_id": 16,
+            "origin_slot": 0,
+            "target_id": 12,
+            "target_slot": 2,
+            "type": "SAMPLER"
+          },
+          {
+            "id": 14,
+            "origin_id": 17,
+            "origin_slot": 0,
+            "target_id": 12,
+            "target_slot": 3,
+            "type": "SIGMAS"
+          },
+          {
+            "id": 15,
+            "origin_id": 11,
+            "origin_slot": 0,
+            "target_id": 12,
+            "target_slot": 4,
+            "type": "LATENT"
+          },
+          {
+            "id": 16,
+            "origin_id": 12,
+            "origin_slot": 0,
+            "target_id": 13,
+            "target_slot": 0,
+            "type": "LATENT"
+          },
+          {
+            "id": 17,
+            "origin_id": 9,
+            "origin_slot": 0,
+            "target_id": 13,
+            "target_slot": 1,
+            "type": "VAE"
+          },
+          {
+            "id": 34,
+            "origin_id": 31,
+            "origin_slot": 1,
+            "target_id": 17,
+            "target_slot": 1,
+            "type": "INT"
+          },
+          {
+            "id": 37,
+            "origin_id": 32,
+            "origin_slot": 1,
+            "target_id": 17,
+            "target_slot": 2,
+            "type": "INT"
+          },
+          {
+            "id": 32,
+            "origin_id": 27,
+            "origin_slot": 0,
+            "target_id": 31,
+            "target_slot": 0,
+            "type": "INT"
+          },
+          {
+            "id": 35,
+            "origin_id": 28,
+            "origin_slot": 0,
+            "target_id": 32,
+            "target_slot": 0,
+            "type": "INT"
+          },
+          {
+            "id": 25,
+            "origin_id": 13,
+            "origin_slot": 0,
+            "target_id": -20,
+            "target_slot": 0,
+            "type": "IMAGE"
+          },
+          {
+            "id": 152,
+            "origin_id": -10,
+            "origin_slot": 0,
+            "target_id": 24,
+            "target_slot": 1,
+            "type": "STRING"
+          },
+          {
+            "id": 153,
+            "origin_id": -10,
+            "origin_slot": 1,
+            "target_id": 27,
+            "target_slot": 0,
+            "type": "INT"
+          },
+          {
+            "id": 154,
+            "origin_id": -10,
+            "origin_slot": 2,
+            "target_id": 28,
+            "target_slot": 0,
+            "type": "INT"
+          },
+          {
+            "id": 155,
+            "origin_id": -10,
+            "origin_slot": 3,
+            "target_id": 18,
+            "target_slot": 0,
+            "type": "INT"
+          },
+          {
+            "id": 156,
+            "origin_id": -10,
+            "origin_slot": 4,
+            "target_id": 23,
+            "target_slot": 0,
+            "type": "COMBO"
+          },
+          {
+            "id": 158,
+            "origin_id": -10,
+            "origin_slot": 6,
+            "target_id": 9,
+            "target_slot": 0,
+            "type": "COMBO"
+          },
+          {
+            "id": 195,
+            "origin_id": 145,
+            "origin_slot": 0,
+            "target_id": 144,
+            "target_slot": 0,
+            "type": "STRING"
+          },
+          {
+            "id": 196,
+            "origin_id": 148,
+            "origin_slot": 0,
+            "target_id": 145,
+            "target_slot": 0,
+            "type": "STRING"
+          },
+          {
+            "id": 197,
+            "origin_id": 150,
+            "origin_slot": 0,
+            "target_id": 146,
+            "target_slot": 0,
+            "type": "STRING"
+          },
+          {
+            "id": 199,
+            "origin_id": 147,
+            "origin_slot": 0,
+            "target_id": 148,
+            "target_slot": 0,
+            "type": "STRING"
+          },
+          {
+            "id": 200,
+            "origin_id": 148,
+            "origin_slot": 0,
+            "target_id": 149,
+            "target_slot": 0,
+            "type": "STRING"
+          },
+          {
+            "id": 201,
+            "origin_id": 148,
+            "origin_slot": 0,
+            "target_id": 150,
+            "target_slot": 0,
+            "type": "STRING"
+          },
+          {
+            "id": 202,
+            "origin_id": 149,
+            "origin_slot": 0,
+            "target_id": 151,
+            "target_slot": 0,
+            "type": "STRING"
+          },
+          {
+            "id": 207,
+            "origin_id": 151,
+            "origin_slot": 1,
+            "target_id": 17,
+            "target_slot": 0,
+            "type": "INT"
+          },
+          {
+            "id": 208,
+            "origin_id": 144,
+            "origin_slot": 0,
+            "target_id": 17,
+            "target_slot": 3,
+            "type": "FLOAT"
+          },
+          {
+            "id": 209,
+            "origin_id": 146,
+            "origin_slot": 0,
+            "target_id": 17,
+            "target_slot": 4,
+            "type": "FLOAT"
+          },
+          {
+            "id": 211,
+            "origin_id": 154,
+            "origin_slot": 0,
+            "target_id": 155,
+            "target_slot": 2,
+            "type": "MODEL"
+          },
+          {
+            "id": 213,
+            "origin_id": 24,
+            "origin_slot": 0,
+            "target_id": 155,
+            "target_slot": 1,
+            "type": "CONDITIONING"
+          },
+          {
+            "id": 214,
+            "origin_id": 10,
+            "origin_slot": 0,
+            "target_id": 155,
+            "target_slot": 3,
+            "type": "CONDITIONING"
+          },
+          {
+            "id": 215,
+            "origin_id": 155,
+            "origin_slot": 0,
+            "target_id": 12,
+            "target_slot": 1,
+            "type": "GUIDER"
+          },
+          {
+            "id": 216,
+            "origin_id": -10,
+            "origin_slot": 7,
+            "target_id": 154,
+            "target_slot": 0,
+            "type": "COMBO"
+          },
+          {
+            "id": 218,
+            "origin_id": 156,
+            "origin_slot": 0,
+            "target_id": 147,
+            "target_slot": 0,
+            "type": "STRING"
+          },
+          {
+            "id": 219,
+            "origin_id": -10,
+            "origin_slot": 8,
+            "target_id": 156,
+            "target_slot": 0,
+            "type": "COMBO"
+          },
+          {
+            "id": 222,
+            "origin_id": 23,
+            "origin_slot": 0,
+            "target_id": 157,
+            "target_slot": 0,
+            "type": "MODEL"
+          },
+          {
+            "id": 223,
+            "origin_id": 157,
+            "origin_slot": 0,
+            "target_id": 155,
+            "target_slot": 0,
+            "type": "MODEL"
+          },
+          {
+            "id": 253,
+            "origin_id": -10,
+            "origin_slot": 5,
+            "target_id": 177,
+            "target_slot": 0,
+            "type": "COMBO"
+          },
+          {
+            "id": 254,
+            "origin_id": 177,
+            "origin_slot": 0,
+            "target_id": 24,
+            "target_slot": 0,
+            "type": "CLIP"
+          }
+        ],
+        "extra": {
+          "ue_links": [],
+          "links_added_by_ue": []
+        }
+      }
+    ]
+  },
+  "config": {},
+  "extra": {
+    "ds": {
+      "scale": 0.7247295000000012,
+      "offset": [
+        -5909.657794514147,
+        -74.13227900158151
+      ]
+    },
+    "frontendVersion": "1.45.15",
+    "VHS_latentpreview": false,
+    "VHS_latentpreviewrate": 0,
+    "VHS_MetadataImage": true,
+    "VHS_KeepIntermediate": true,
+    "ue_links": [],
+    "links_added_by_ue": []
+  },
+  "version": 0.4
+}
\ No newline at end of file
diff --git a/ideogram_prompt_builder.py b/ideogram_prompt_builder.py
new file mode 100644
index 0000000..6a4f3e3
--- /dev/null
+++ b/ideogram_prompt_builder.py
@@ -0,0 +1,3010 @@
+import copy
+import json
+import re
+import shutil
+import sys
+import time
+import urllib.error
+import urllib.parse
+import urllib.request
+import uuid
+import zipfile
+from datetime import datetime
+from pathlib import Path
+
+from PyQt6.QtCore import QPointF, QRectF, QThread, Qt, pyqtSignal
+from PyQt6.QtGui import QAction, QColor, QGuiApplication, QKeySequence, QPainter, QPen, QPixmap
+from PyQt6.QtWidgets import (
+    QApplication,
+    QCheckBox,
+    QColorDialog,
+    QComboBox,
+    QDialog,
+    QDialogButtonBox,
+    QFileDialog,
+    QFormLayout,
+    QFrame,
+    QGridLayout,
+    QGroupBox,
+    QHBoxLayout,
+    QLabel,
+    QLineEdit,
+    QListWidget,
+    QListWidgetItem,
+    QMainWindow,
+    QMessageBox,
+    QPlainTextEdit,
+    QProgressDialog,
+    QPushButton,
+    QRadioButton,
+    QScrollArea,
+    QSizePolicy,
+    QSlider,
+    QSpinBox,
+    QSplitter,
+    QTabWidget,
+    QTextEdit,
+    QToolButton,
+    QVBoxLayout,
+    QWidget,
+    QInputDialog,
+)
+
+
+HEX_RE = re.compile(r"^#[0-9A-F]{6}$")
+MIN_BBOX_SIZE = 16
+APP_DIR = Path(__file__).resolve().parent
+LIBRARY_FILE = APP_DIR / "prompt_library.json"
+PREVIEW_DIR = APP_DIR / "prompt_previews"
+LANG_FILE = APP_DIR / "translations.json"
+SETTINGS_FILE = APP_DIR / "comfy_settings.json"
+DRAFT_FILE = APP_DIR / "draft.json"
+WORKFLOW_FILE = APP_DIR / "ideogram4NSFWComfyui_v11.json"
+DEFAULT_LANGUAGE = "en"
+LANGUAGE_NAMES = {"en": "English", "ru": "Русский"}
+MAX_UNDO = 60
+
+DEFAULT_SETTINGS = {
+    "comfy_host": "127.0.0.1",
+    "comfy_port": 8188,
+    "comfy_https": False,
+    "theme": "light",
+    "language": DEFAULT_LANGUAGE,
+}
+
+# Models / samplers / custom nodes the bundled workflow needs to run in ComfyUI.
+REQUIRED_COMFY = {
+    "nodes": [
+        "Ideogram4PromptBuilderKJ",
+        "Ideogram4Scheduler",
+        "UNETLoader",
+        "VAELoader",
+        "CLIPLoader",
+        "CLIPLoaderGGUF",
+        "DualModelGuider",
+        "SamplerCustomAdvanced",
+        "KSamplerSelect",
+        "RandomNoise",
+        "VAEDecode",
+    ],
+    "unet": [
+        "ideogram4_fp8_scaled.safetensors",
+        "ideogram4_unconditional_fp8_scaled.safetensors",
+    ],
+    "vae": ["flux2-vae.safetensors"],
+    "clip": ["qwen3vl_8b_fp8_scaled.safetensors"],
+    "clip_gguf": ["Qwen3VL-8B-Uncensored-HauhauCS-Aggressive-Q8_0.gguf"],
+    "samplers": ["euler"],
+}
+
+# Quick-insert element templates (item 12).
+ELEMENT_TEMPLATES = {
+    "Character": {
+        "type": "obj",
+        "desc": "A full-body character with realistic proportions and natural posture.",
+        "bbox": [120, 320, 950, 690],
+    },
+    "Title text": {
+        "type": "text",
+        "text": "TITLE",
+        "desc": "Bold display lettering across the top of the composition.",
+        "bbox": [70, 120, 200, 880],
+    },
+    "Background object": {
+        "type": "obj",
+        "desc": "A secondary object that anchors the background of the scene.",
+        "bbox": [400, 100, 800, 500],
+    },
+}
+
+THEMES = {
+    "light": {
+        "bg": "#F4F6F4", "panel": "#FFFFFF", "text": "#182024", "muted": "#5D666F",
+        "border": "#DDE3DD", "field_border": "#CBD5CE", "accent": "#176B87",
+        "accent_dark": "#0F5269", "list_sel_bg": "#DCEFF3", "list_sel_fg": "#0F5269",
+        "hover_bg": "#F0F7F8", "canvas_bg": "#F3F6F3", "canvas_grid": "#D4DCD4",
+        "canvas_label": "#98A39B", "error": "#C0392B",
+    },
+    "dark": {
+        "bg": "#1E2227", "panel": "#272C33", "text": "#E6EAED", "muted": "#9AA4AD",
+        "border": "#363C44", "field_border": "#3C434C", "accent": "#3AA6C4",
+        "accent_dark": "#2C8BA6", "list_sel_bg": "#234049", "list_sel_fg": "#CDEBF3",
+        "hover_bg": "#2E3942", "canvas_bg": "#22272D", "canvas_grid": "#3A424A",
+        "canvas_label": "#6C757D", "error": "#E06A5C",
+    },
+}
+
+
+def build_stylesheet(theme):
+    c = THEMES.get(theme, THEMES["light"])
+    return f"""
+        QMainWindow, QWidget {{ background: {c['bg']}; color: {c['text']}; font-family: Segoe UI; font-size: 10.5pt; }}
+        QToolBar {{ background: {c['panel']}; border: 0; border-bottom: 1px solid {c['border']}; spacing: 8px; padding: 8px; }}
+        QToolButton {{ border-radius: 7px; padding: 6px 10px; color: {c['text']}; }}
+        QToolBar QToolButton:hover {{ background: {c['hover_bg']}; }}
+        QGroupBox {{ background: {c['panel']}; border: 1px solid {c['border']}; border-radius: 10px; margin-top: 18px; padding: 14px; font-weight: 700; }}
+        QGroupBox::title {{ subcontrol-origin: margin; left: 14px; padding: 0 7px; color: {c['accent']}; }}
+        QLineEdit, QTextEdit, QPlainTextEdit, QComboBox, QSpinBox, QListWidget {{
+            background: {c['panel']}; border: 1px solid {c['field_border']}; border-radius: 8px; padding: 7px;
+            selection-background-color: {c['accent']}; color: {c['text']};
+        }}
+        QLineEdit:focus, QTextEdit:focus, QPlainTextEdit:focus, QComboBox:focus, QSpinBox:focus {{
+            border: 1px solid {c['accent']};
+        }}
+        QLineEdit[invalid="true"], QPlainTextEdit[invalid="true"] {{ border: 1px solid {c['error']}; }}
+        QPushButton {{
+            background: {c['panel']}; border: 1px solid {c['field_border']}; border-radius: 8px; padding: 8px 12px; color: {c['text']};
+        }}
+        QPushButton:hover {{ border-color: {c['accent']}; background: {c['hover_bg']}; }}
+        QPushButton:disabled {{ color: {c['muted']}; }}
+        QPushButton#PrimaryButton {{ background: {c['accent']}; color: white; border-color: {c['accent']}; font-weight: 700; }}
+        QPushButton#PrimaryButton:hover {{ background: {c['accent_dark']}; }}
+        QListWidget::item {{ padding: 8px; border-radius: 6px; }}
+        QListWidget::item:selected {{ background: {c['list_sel_bg']}; color: {c['list_sel_fg']}; }}
+        QRadioButton, QCheckBox {{ background: transparent; }}
+        QLabel {{ background: transparent; }}
+    """
+
+
+def load_settings():
+    settings = dict(DEFAULT_SETTINGS)
+    if SETTINGS_FILE.exists():
+        try:
+            with open(SETTINGS_FILE, "r", encoding="utf-8") as handle:
+                data = json.load(handle)
+            if isinstance(data, dict):
+                settings.update({k: data[k] for k in DEFAULT_SETTINGS if k in data})
+        except (OSError, json.JSONDecodeError):
+            pass
+    return settings
+
+
+def save_settings(settings):
+    try:
+        with open(SETTINGS_FILE, "w", encoding="utf-8") as handle:
+            json.dump(settings, handle, ensure_ascii=False, indent=2)
+    except OSError:
+        pass
+
+
+DEFAULT_TRANSLATIONS = {
+    "en": {
+        "app.title": "Ideogram 4 Prompt Builder",
+        "tb.example": "Example",
+        "tb.import": "Import JSON",
+        "tb.save_json": "Save JSON",
+        "tb.copy": "Copy",
+        "tb.save_library": "Save to library",
+        "tb.library": "Library",
+        "tb.language": "Language:",
+        "grp.high": "High level description",
+        "high.placeholder": "One- or two-sentence summary of the full image...",
+        "grp.presets": "Presets",
+        "preset.no_safety": "Add \"No safety filter.\"",
+        "grp.style": "Style",
+        "style.photo": "Photo",
+        "style.art": "Art",
+        "style.aesthetics": "Aesthetics",
+        "style.lighting": "Lighting",
+        "style.photo_field": "Photo",
+        "style.art_style": "Art style",
+        "style.medium": "Medium",
+        "style.palette": "Palette",
+        "grp.composition": "Composition",
+        "comp.background": "Background",
+        "comp.background_placeholder": "Describe the environment or background before listing foreground elements...",
+        "comp.add_element": "Add element",
+        "comp.remove_element": "Remove element",
+        "comp.hint": "Drag the rectangle to move the element. Drag the round handles to resize the bbox.",
+        "el.type": "Type",
+        "el.label": "Label",
+        "el.text": "Text",
+        "el.description": "Description",
+        "el.palette": "Palette",
+        "el.use_bbox": "Use bbox",
+        "el.bbox": "BBox",
+        "el.element": "Element",
+        "out.title": "Ready JSON",
+        "out.pretty": "Pretty",
+        "out.compact": "Compact",
+        "out.copy_compact": "Copy compact",
+        "out.save_json_btn": "Save .json",
+        "canvas.label": "bbox canvas 0-1000",
+        "val.ok": "JSON assembled in Ideogram 4 key order and ready for ComfyUI.",
+        "val.no_high": "Add high_level_description for better scene adherence.",
+        "val.bg_required": "background is required.",
+        "val.add_element": "Add at least one element.",
+        "val.style_missing": "style_description is missing: {fields}.",
+        "val.photo_or_art": "Exactly one key required: photo or art_style.",
+        "val.hex_upper": "Color {color} must be uppercase #RRGGBB.",
+        "val.text_literal": "{title}: text element requires a literal text.",
+        "val.desc_required": "{title}: desc is required.",
+        "val.bbox_order": "{title}: bbox must have y_max/x_max greater than y_min/x_min.",
+        "val.el_hex": "{title}: color {color} must be uppercase #RRGGBB.",
+        "val.element_word": "element {index}",
+        "pal.placeholder": "#1E73BE, #FDFDFD",
+        "pal.add": "Add color",
+        "pal.configure": "Configure color",
+        "pal.swatch_tip": "{color}: click to configure",
+        "pal.remove": "Remove color",
+        "dlg.save_json_title": "Save JSON",
+        "dlg.json_filter": "JSON files (*.json);;All files (*)",
+        "dlg.import_title": "Import JSON",
+        "imp.error_title": "Import error",
+        "trn.error_title": "Translate error",
+        "trn.error_msg": "Translation failed:\n{err}",
+        "trn.to_ru": "Translate to RU",
+        "trn.to_en": "Translate to EN",
+        "lib.name_prompt": "Prompt name:",
+        "lib.untitled": "Untitled",
+        "lib.preview_q_title": "Preview",
+        "lib.preview_q": "Attach a preview image to this prompt?",
+        "lib.save_fail": "Failed to save:\n{err}",
+        "lib.saved": "Prompt \"{name}\" saved.",
+        "prev.pick_title": "Choose preview image",
+        "prev.filter": "Images (*.png *.jpg *.jpeg *.webp *.bmp);;All files (*)",
+        "prev.save_fail": "Failed to save image:\n{err}",
+        "prev.title": "Preview",
+        "libd.title": "Prompt library",
+        "libd.saved_prompts": "Saved prompts",
+        "libd.no_preview": "No preview",
+        "libd.preview_unavailable": "Preview unavailable",
+        "libd.use": "Load into editor",
+        "libd.rename": "Rename",
+        "libd.set_preview": "Set preview",
+        "libd.clear_preview": "Clear preview",
+        "libd.delete": "Delete",
+        "libd.close": "Close",
+        "libd.rename_title": "Rename",
+        "libd.rename_label": "Name:",
+        "libd.delete_title": "Delete prompt",
+        "libd.delete_q": "Delete \"{name}\" from the library?",
+        "libd.meta": "Updated: {updated}\nElements: {count}\n\n{high}",
+        "libd.no_high": "(no high_level_description)",
+        "libd.search": "Search...",
+        "libd.tags": "Tags (comma-separated):",
+        "libd.tags_col": "Tags",
+        "libd.paste_preview": "Paste preview from clipboard",
+        "libd.export": "Export library...",
+        "libd.import": "Import library...",
+        "libd.no_clipboard_image": "No image in the clipboard.",
+        "libd.export_done": "Library exported to:\n{path}",
+        "libd.import_done": "Imported {count} prompt(s).",
+        "libd.export_fail": "Export failed:\n{err}",
+        "libd.import_fail": "Import failed:\n{err}",
+        "libd.export_filter": "ZIP archive (*.zip)",
+        "tb.undo": "Undo",
+        "tb.redo": "Redo",
+        "tb.duplicate": "Duplicate element",
+        "tb.move_up": "Move up",
+        "tb.move_down": "Move down",
+        "tb.theme": "Theme",
+        "tb.comfy_settings": "ComfyUI settings",
+        "tb.generate": "Generate in ComfyUI",
+        "tb.check_comfy": "Check ComfyUI",
+        "tb.template": "Add from template",
+        "tb.overwrite": "Update in library",
+        "menu.file": "File",
+        "menu.edit": "Edit",
+        "menu.library": "Library",
+        "menu.comfy": "ComfyUI",
+        "menu.view": "View",
+        "canvas.load_ref": "Reference image...",
+        "canvas.paste_ref": "Paste reference",
+        "canvas.clear_ref": "Clear reference",
+        "canvas.zoom": "Grid scale",
+        "canvas.ref_load_fail": "Could not load image.",
+        "counter.colors": "{count}/{limit} colors",
+        "set.title": "ComfyUI connection settings",
+        "set.host": "Host:",
+        "set.port": "Port:",
+        "set.https": "Use HTTPS",
+        "set.test": "Test connection",
+        "set.test_ok": "Connection OK. ComfyUI is reachable.",
+        "set.test_fail": "Connection failed:\n{err}",
+        "set.saved": "Settings saved.",
+        "comfy.checking": "Checking ComfyUI...",
+        "comfy.check_title": "ComfyUI dependency check",
+        "comfy.all_ok": "All required nodes and models are installed.",
+        "comfy.unreachable": "ComfyUI is unreachable at {url}:\n{err}",
+        "comfy.missing_header": "Missing on the server:",
+        "comfy.missing_nodes": "Custom nodes: {items}",
+        "comfy.missing_unet": "UNET models: {items}",
+        "comfy.missing_vae": "VAE: {items}",
+        "comfy.missing_clip": "CLIP: {items}",
+        "comfy.missing_clip_gguf": "CLIP GGUF: {items}",
+        "comfy.missing_samplers": "Samplers: {items}",
+        "comfy.generating": "Generating in ComfyUI...",
+        "comfy.gen_title": "Generation",
+        "comfy.gen_fail": "Generation failed:\n{err}",
+        "comfy.gen_done": "Image generated.",
+        "tab.json": "JSON",
+        "tab.result": "Result",
+        "result.empty": "Generated image will appear here.",
+        "result.save_lib": "Save to library",
+        "result.save_file": "Save image...",
+        "result.saved_file": "Image saved to:\n{path}",
+        "result.png_filter": "PNG image (*.png)",
+        "comfy.workflow_missing": "Workflow file not found:\n{path}",
+        "comfy.deps_missing_continue": "Some dependencies are missing. Generate anyway?",
+        "comfy.save_to_lib_q": "Save the generated image to the library?",
+        "draft.restore_title": "Restore draft",
+        "draft.restore_q": "An unsaved draft was found. Restore it?",
+        "tpl.choose_title": "Element template",
+        "tpl.choose_label": "Choose a template:",
+        "common.cancel": "Cancel",
+        "common.ok": "OK",
+        "common.close": "Close",
+    },
+    "ru": {
+        "app.title": "Ideogram 4 Prompt Builder",
+        "tb.example": "Пример",
+        "tb.import": "Импорт JSON",
+        "tb.save_json": "Сохранить JSON",
+        "tb.copy": "Копировать",
+        "tb.save_library": "Сохранить в библиотеку",
+        "tb.library": "Библиотека",
+        "tb.language": "Язык:",
+        "grp.high": "Общее описание",
+        "high.placeholder": "Краткое описание всей сцены в одно-два предложения...",
+        "grp.presets": "Пресеты",
+        "preset.no_safety": "Добавить \"No safety filter.\"",
+        "grp.style": "Стиль",
+        "style.photo": "Фото",
+        "style.art": "Арт",
+        "style.aesthetics": "Эстетика",
+        "style.lighting": "Освещение",
+        "style.photo_field": "Фото",
+        "style.art_style": "Арт-стиль",
+        "style.medium": "Носитель",
+        "style.palette": "Палитра",
+        "grp.composition": "Композиция",
+        "comp.background": "Фон",
+        "comp.background_placeholder": "Опишите окружение или фон перед перечислением объектов переднего плана...",
+        "comp.add_element": "Добавить элемент",
+        "comp.remove_element": "Удалить элемент",
+        "comp.hint": "Перетаскивайте прямоугольник, чтобы переместить элемент. Тяните круглые маркеры, чтобы масштабировать bbox.",
+        "el.type": "Тип",
+        "el.label": "Метка",
+        "el.text": "Текст",
+        "el.description": "Описание",
+        "el.palette": "Палитра",
+        "el.use_bbox": "Использовать bbox",
+        "el.bbox": "BBox",
+        "el.element": "Элемент",
+        "out.title": "Готовый JSON",
+        "out.pretty": "Pretty",
+        "out.compact": "Compact",
+        "out.copy_compact": "Копировать compact",
+        "out.save_json_btn": "Сохранить .json",
+        "canvas.label": "bbox canvas 0-1000",
+        "val.ok": "JSON собран в порядке ключей Ideogram 4 и готов для ComfyUI.",
+        "val.no_high": "Добавьте high_level_description для лучшего следования сцене.",
+        "val.bg_required": "background обязателен.",
+        "val.add_element": "Добавьте хотя бы один элемент.",
+        "val.style_missing": "В style_description не хватает: {fields}.",
+        "val.photo_or_art": "Нужен ровно один ключ: photo или art_style.",
+        "val.hex_upper": "Цвет {color} должен быть uppercase #RRGGBB.",
+        "val.text_literal": "{title}: для text-элемента нужен literal text.",
+        "val.desc_required": "{title}: desc обязателен.",
+        "val.bbox_order": "{title}: bbox должен иметь y_max/x_max больше y_min/x_min.",
+        "val.el_hex": "{title}: цвет {color} должен быть uppercase #RRGGBB.",
+        "val.element_word": "element {index}",
+        "pal.placeholder": "#1E73BE, #FDFDFD",
+        "pal.add": "Добавить цвет",
+        "pal.configure": "Настроить цвет",
+        "pal.swatch_tip": "{color}: нажмите, чтобы настроить",
+        "pal.remove": "Удалить цвет",
+        "dlg.save_json_title": "Сохранить JSON",
+        "dlg.json_filter": "JSON файлы (*.json);;Все файлы (*)",
+        "dlg.import_title": "Импорт JSON",
+        "imp.error_title": "Ошибка импорта",
+        "trn.error_title": "Ошибка перевода",
+        "trn.error_msg": "Не удалось выполнить перевод:\n{err}",
+        "trn.to_ru": "Перевести на RU",
+        "trn.to_en": "Перевести на EN",
+        "lib.name_prompt": "Название промта:",
+        "lib.untitled": "Без названия",
+        "lib.preview_q_title": "Превью",
+        "lib.preview_q": "Добавить изображение превью к этому промту?",
+        "lib.save_fail": "Не удалось сохранить:\n{err}",
+        "lib.saved": "Промт «{name}» сохранён.",
+        "prev.pick_title": "Выбрать изображение превью",
+        "prev.filter": "Изображения (*.png *.jpg *.jpeg *.webp *.bmp);;Все файлы (*)",
+        "prev.save_fail": "Не удалось сохранить изображение:\n{err}",
+        "prev.title": "Превью",
+        "libd.title": "Библиотека промтов",
+        "libd.saved_prompts": "Сохранённые промты",
+        "libd.no_preview": "Нет превью",
+        "libd.preview_unavailable": "Превью недоступно",
+        "libd.use": "Загрузить в редактор",
+        "libd.rename": "Переименовать",
+        "libd.set_preview": "Задать превью",
+        "libd.clear_preview": "Убрать превью",
+        "libd.delete": "Удалить",
+        "libd.close": "Закрыть",
+        "libd.rename_title": "Переименовать",
+        "libd.rename_label": "Название:",
+        "libd.delete_title": "Удалить промт",
+        "libd.delete_q": "Удалить «{name}» из библиотеки?",
+        "libd.meta": "Обновлено: {updated}\nЭлементов: {count}\n\n{high}",
+        "libd.no_high": "(без high_level_description)",
+        "libd.search": "Поиск...",
+        "libd.tags": "Теги (через запятую):",
+        "libd.tags_col": "Теги",
+        "libd.paste_preview": "Вставить превью из буфера",
+        "libd.export": "Экспорт библиотеки...",
+        "libd.import": "Импорт библиотеки...",
+        "libd.no_clipboard_image": "В буфере обмена нет изображения.",
+        "libd.export_done": "Библиотека экспортирована в:\n{path}",
+        "libd.import_done": "Импортировано промтов: {count}.",
+        "libd.export_fail": "Не удалось экспортировать:\n{err}",
+        "libd.import_fail": "Не удалось импортировать:\n{err}",
+        "libd.export_filter": "ZIP архив (*.zip)",
+        "tb.undo": "Отменить",
+        "tb.redo": "Повторить",
+        "tb.duplicate": "Дублировать элемент",
+        "tb.move_up": "Вверх",
+        "tb.move_down": "Вниз",
+        "tb.theme": "Тема",
+        "tb.comfy_settings": "Настройки ComfyUI",
+        "tb.generate": "Сгенерировать в ComfyUI",
+        "tb.check_comfy": "Проверить ComfyUI",
+        "tb.template": "Добавить из шаблона",
+        "tb.overwrite": "Обновить в библиотеке",
+        "menu.file": "Файл",
+        "menu.edit": "Правка",
+        "menu.library": "Библиотека",
+        "menu.comfy": "ComfyUI",
+        "menu.view": "Вид",
+        "canvas.load_ref": "Референс-изображение...",
+        "canvas.paste_ref": "Вставить референс",
+        "canvas.clear_ref": "Убрать референс",
+        "canvas.zoom": "Масштаб сетки",
+        "canvas.ref_load_fail": "Не удалось загрузить изображение.",
+        "counter.colors": "{count}/{limit} цветов",
+        "set.title": "Настройки соединения с ComfyUI",
+        "set.host": "Хост:",
+        "set.port": "Порт:",
+        "set.https": "Использовать HTTPS",
+        "set.test": "Проверить соединение",
+        "set.test_ok": "Соединение установлено. ComfyUI доступен.",
+        "set.test_fail": "Не удалось подключиться:\n{err}",
+        "set.saved": "Настройки сохранены.",
+        "comfy.checking": "Проверка ComfyUI...",
+        "comfy.check_title": "Проверка зависимостей ComfyUI",
+        "comfy.all_ok": "Все необходимые ноды и модели установлены.",
+        "comfy.unreachable": "ComfyUI недоступен по адресу {url}:\n{err}",
+        "comfy.missing_header": "Отсутствует на сервере:",
+        "comfy.missing_nodes": "Кастомные ноды: {items}",
+        "comfy.missing_unet": "UNET-модели: {items}",
+        "comfy.missing_vae": "VAE: {items}",
+        "comfy.missing_clip": "CLIP: {items}",
+        "comfy.missing_clip_gguf": "CLIP GGUF: {items}",
+        "comfy.missing_samplers": "Семплеры: {items}",
+        "comfy.generating": "Генерация в ComfyUI...",
+        "comfy.gen_title": "Генерация",
+        "comfy.gen_fail": "Не удалось сгенерировать:\n{err}",
+        "comfy.gen_done": "Изображение сгенерировано.",
+        "tab.json": "JSON",
+        "tab.result": "Результат",
+        "result.empty": "Здесь появится сгенерированное изображение.",
+        "result.save_lib": "Сохранить в библиотеку",
+        "result.save_file": "Сохранить изображение...",
+        "result.saved_file": "Изображение сохранено в:\n{path}",
+        "result.png_filter": "PNG изображение (*.png)",
+        "comfy.workflow_missing": "Файл workflow не найден:\n{path}",
+        "comfy.deps_missing_continue": "Некоторые зависимости отсутствуют. Всё равно сгенерировать?",
+        "comfy.save_to_lib_q": "Сохранить сгенерированное изображение в библиотеку?",
+        "draft.restore_title": "Восстановить черновик",
+        "draft.restore_q": "Найден несохранённый черновик. Восстановить его?",
+        "tpl.choose_title": "Шаблон элемента",
+        "tpl.choose_label": "Выберите шаблон:",
+        "common.cancel": "Отмена",
+        "common.ok": "ОК",
+        "common.close": "Закрыть",
+    },
+}
+
+
+def ensure_translation_file():
+    """Write the bundled translations to disk if no file exists yet."""
+    if not LANG_FILE.exists():
+        try:
+            with open(LANG_FILE, "w", encoding="utf-8") as handle:
+                json.dump(DEFAULT_TRANSLATIONS, handle, ensure_ascii=False, indent=2)
+        except OSError:
+            pass
+
+
+def load_translations():
+    """Load translations from the external file, falling back to bundled defaults."""
+    ensure_translation_file()
+    try:
+        with open(LANG_FILE, "r", encoding="utf-8") as handle:
+            data = json.load(handle)
+    except (OSError, json.JSONDecodeError):
+        data = {}
+    if not isinstance(data, dict) or not data:
+        return {lang: dict(strings) for lang, strings in DEFAULT_TRANSLATIONS.items()}
+    return data
+
+
+TRANSLATIONS = load_translations()
+_saved_lang = load_settings().get("language", DEFAULT_LANGUAGE)
+if _saved_lang in TRANSLATIONS:
+    CURRENT_LANG = _saved_lang
+elif DEFAULT_LANGUAGE in TRANSLATIONS:
+    CURRENT_LANG = DEFAULT_LANGUAGE
+else:
+    CURRENT_LANG = next(iter(TRANSLATIONS))
+
+
+def available_languages():
+    return list(TRANSLATIONS.keys())
+
+
+def tr(key):
+    """Translate a key into the current language, falling back to English then the key."""
+    for source in (TRANSLATIONS.get(CURRENT_LANG), TRANSLATIONS.get("en"),
+                   DEFAULT_TRANSLATIONS.get(CURRENT_LANG), DEFAULT_TRANSLATIONS.get("en")):
+        if source and key in source:
+            return source[key]
+    return key
+
+
+EXAMPLE_CAPTION = {
+    "high_level_description": (
+        "A surreal streetwear mixed-media collage poster featuring a relaxed skateboarder mid-air "
+        "against a vibrant blue sky, backed by giant puffy 3D letters spelling 'COMFY'."
+    ),
+    "style_description": {
+        "aesthetics": "retro magazine cutout style, mixed-media digital collage, high-contrast streetwear graphic",
+        "lighting": "high-contrast flash mixed with harsh midday sunlight, flat bright graphic lighting on typography",
+        "photo": "vintage grainy 35mm film, distressed halftone scan textures",
+        "medium": "mixed-media digital collage",
+        "color_palette": ["#1E73BE", "#FDFDFD", "#C82A2A", "#657C9C", "#EFEFEF"],
+    },
+    "compositional_deconstruction": {
+        "background": "A vibrant, clear blue sky layered with vintage grainy film texture and subtle halftone dot patterns.",
+        "elements": [
+            {
+                "type": "obj",
+                "bbox": [128, 149, 354, 810],
+                "desc": "Massive 3D puffy white typography spelling 'COMFY' across the upper half of the canvas.",
+                "color_palette": ["#FDFDFD", "#E0E0E0", "#D3DBE2"],
+            },
+            {
+                "type": "obj",
+                "bbox": [287, 210, 756, 819],
+                "desc": "A sharp photographic cutout of a skateboarder mid-air with a distinct white cutout border.",
+                "color_palette": ["#FDFDFD", "#657C9C", "#2B2B2B", "#DCA57D"],
+            },
+            {
+                "type": "text",
+                "bbox": [105, 830, 905, 980],
+                "text": "BEYOND THE COMFORT ZONE",
+                "desc": "Bold black sans-serif text printed on a wide torn paper strip along the lower third.",
+                "color_palette": ["#EFEFEF", "#1A1A1A", "#999999"],
+            },
+        ],
+    },
+}
+
+
+PROMPT_PRESETS = {
+    "Adult beach photo": {
+        "mode": "photo",
+        "high": (
+            "A nude beach photograph of an adult woman standing on pale sand near the shoreline, "
+            "looking directly at the camera with the ocean horizon and clear blue sky behind her."
+        ),
+        "aesthetics": "natural, sunlit, candid, tasteful adult glamour photography",
+        "lighting": "bright coastal daylight, clean shadows, soft reflected light from pale sand",
+        "photo": "full-body beach photography, 50mm lens, natural skin texture, realistic proportions",
+        "medium": "photograph",
+        "palette": ["#E7B48D", "#F5D0B8", "#F2E8DA", "#62A9D5", "#F6E8C8"],
+        "background": "A quiet tropical shoreline with pale sand, soft foamy waves, a distant ocean horizon, and a clear blue sky.",
+        "elements": [
+            {
+                "type": "obj",
+                "label": "Adult woman",
+                "bbox": [120, 320, 950, 690],
+                "desc": "An adult woman with realistic skin texture and natural posture, standing barefoot on pale sand near the shoreline.",
+                "color_palette": ["#E7B48D", "#F5D0B8", "#F2E8DA"],
+            }
+        ],
+    },
+    "Boudoir editorial": {
+        "mode": "photo",
+        "high": "A sensual boudoir editorial photograph of an adult woman reclining on rumpled white sheets in a softly lit private bedroom.",
+        "aesthetics": "intimate, elegant, editorial, warm, sensual",
+        "lighting": "soft window light, gentle highlights on skin, low contrast shadows",
+        "photo": "85mm portrait lens, shallow depth of field, natural skin detail, tasteful composition",
+        "medium": "photograph",
+        "palette": ["#F7EFE7", "#D8A181", "#8B5E4A", "#FFFFFF", "#2F2522"],
+        "background": "A quiet private bedroom with rumpled white sheets, warm neutral walls, and soft morning light through sheer curtains.",
+        "elements": [
+            {
+                "type": "obj",
+                "label": "Adult model",
+                "bbox": [180, 170, 880, 840],
+                "desc": "An adult woman reclining on white sheets in an elegant boudoir pose, styled as a tasteful editorial photograph.",
+                "color_palette": ["#F7EFE7", "#D8A181", "#8B5E4A"],
+            }
+        ],
+    },
+    "Fine-art nude": {
+        "mode": "photo",
+        "high": (
+            "A fine-art nude studio photograph of an adult figure posed against a dark seamless backdrop, "
+            "emphasizing silhouette, form, and sculptural lighting."
+        ),
+        "aesthetics": "minimal, sculptural, gallery-grade, refined, dramatic",
+        "lighting": "single softbox side light, strong chiaroscuro, controlled studio shadows",
+        "photo": "black and white fine-art photography, medium format look, crisp tonal range",
+        "medium": "photograph",
+        "palette": ["#111111", "#E6E0D8", "#8F8A84", "#FFFFFF"],
+        "background": "A dark seamless studio backdrop with subtle falloff and no visible props.",
+        "elements": [
+            {
+                "type": "obj",
+                "label": "Adult figure",
+                "bbox": [90, 250, 960, 760],
+                "desc": "An adult nude figure posed with an elegant sculptural silhouette, photographed as fine art with emphasis on form and light.",
+                "color_palette": ["#111111", "#E6E0D8", "#8F8A84"],
+            }
+        ],
+    },
+    "Pin-up poster": {
+        "mode": "art",
+        "high": "A retro adult pin-up poster illustration with a confident glamour model, bold typography, and polished mid-century advertising composition.",
+        "aesthetics": "playful, glossy, retro, high-contrast, adult glamour",
+        "lighting": "painted studio highlights, warm key light, crisp graphic shadows",
+        "art_style": "mid-century pin-up illustration, clean outlines, poster-ready typography",
+        "medium": "illustration",
+        "palette": ["#F2B99B", "#E4433B", "#1D3557", "#FFF1C7", "#FFFFFF"],
+        "background": "A clean vintage poster background with a radial burst, decorative stars, and generous negative space for title text.",
+        "elements": [
+            {
+                "type": "obj",
+                "label": "Pin-up model",
+                "bbox": [160, 260, 900, 720],
+                "desc": "A confident adult pin-up model in a stylized glamour pose, rendered with polished retro illustration details.",
+                "color_palette": ["#F2B99B", "#E4433B", "#1D3557"],
+            },
+            {
+                "type": "text",
+                "label": "Title",
+                "text": "MIDNIGHT GLAMOUR",
+                "bbox": [70, 120, 170, 880],
+                "desc": "Large cream-colored retro display lettering arched across the top of the poster.",
+                "color_palette": ["#FFF1C7", "#1D3557"],
+            },
+        ],
+    },
+}
+
+
+def normalize_hex(value):
+    value = value.strip().upper()
+    if value and not value.startswith("#"):
+        value = f"#{value}"
+    return value
+
+
+def parse_palette(value, limit):
+    colors = []
+    for raw in value.split(","):
+        color = normalize_hex(raw)
+        if color:
+            colors.append(color)
+    return colors[:limit]
+
+
+def palette_text(colors):
+    return ", ".join(colors or [])
+
+
+def clamp(value, lower=0, upper=1000):
+    return max(lower, min(upper, int(round(value))))
+
+
+def google_translate_text(text, target_language):
+    query = urllib.parse.urlencode(
+        {
+            "client": "gtx",
+            "sl": "auto",
+            "tl": target_language,
+            "dt": "t",
+            "q": text,
+        }
+    )
+    request = urllib.request.Request(
+        f"https://translate.googleapis.com/translate_a/single?{query}",
+        headers={"User-Agent": "Mozilla/5.0"},
+    )
+    with urllib.request.urlopen(request, timeout=12) as response:
+        payload = json.loads(response.read().decode("utf-8"))
+    return "".join(part[0] for part in payload[0] if part and part[0]).strip()
+
+
+_TRANSLATION_CACHE = {}
+
+
+def cached_translate(text, target_language):
+    """Translate with an in-memory cache (item 13) to avoid repeat network calls."""
+    key = (target_language, text)
+    if key in _TRANSLATION_CACHE:
+        return _TRANSLATION_CACHE[key]
+    result = google_translate_text(text, target_language)
+    if result:
+        _TRANSLATION_CACHE[key] = result
+    return result
+
+
+# --- ComfyUI integration --------------------------------------------------
+
+class ComfyError(Exception):
+    pass
+
+
+def comfy_base_url(settings):
+    scheme = "https" if settings.get("comfy_https") else "http"
+    return f"{scheme}://{settings.get('comfy_host', '127.0.0.1')}:{settings.get('comfy_port', 8188)}"
+
+
+def comfy_get(settings, path, timeout=10):
+    url = f"{comfy_base_url(settings)}{path}"
+    request = urllib.request.Request(url, headers={"User-Agent": "IdeogramPromptBuilder"})
+    with urllib.request.urlopen(request, timeout=timeout) as response:
+        return json.loads(response.read().decode("utf-8"))
+
+
+def comfy_post(settings, path, payload, timeout=15):
+    url = f"{comfy_base_url(settings)}{path}"
+    data = json.dumps(payload).encode("utf-8")
+    request = urllib.request.Request(
+        url, data=data, headers={"Content-Type": "application/json", "User-Agent": "IdeogramPromptBuilder"}
+    )
+    with urllib.request.urlopen(request, timeout=timeout) as response:
+        return json.loads(response.read().decode("utf-8"))
+
+
+def comfy_test_connection(settings):
+    """Raise ComfyError if the server is not reachable, else return system stats."""
+    try:
+        return comfy_get(settings, "/system_stats", timeout=6)
+    except (urllib.error.URLError, TimeoutError, OSError, json.JSONDecodeError) as error:
+        raise ComfyError(str(error))
+
+
+def comfy_object_info(settings):
+    try:
+        return comfy_get(settings, "/object_info", timeout=20)
+    except (urllib.error.URLError, TimeoutError, OSError, json.JSONDecodeError) as error:
+        raise ComfyError(str(error))
+
+
+def _combo_values(object_info, node, input_name):
+    """Return the list of allowed values for a combo input of a node, or []."""
+    try:
+        spec = object_info[node]["input"]
+        for section in ("required", "optional"):
+            if input_name in spec.get(section, {}):
+                values = spec[section][input_name][0]
+                return values if isinstance(values, list) else []
+    except (KeyError, TypeError, IndexError):
+        pass
+    return []
+
+
+def check_comfy_dependencies(settings):
+    """Compare REQUIRED_COMFY against a live server. Returns dict of missing items."""
+    info = comfy_object_info(settings)
+    available_nodes = set(info.keys())
+    missing = {
+        "nodes": [n for n in REQUIRED_COMFY["nodes"] if n not in available_nodes],
+        "unet": [], "vae": [], "clip": [], "clip_gguf": [], "samplers": [],
+    }
+
+    unet_values = set(_combo_values(info, "UNETLoader", "unet_name"))
+    for model in REQUIRED_COMFY["unet"]:
+        if unet_values and model not in unet_values:
+            missing["unet"].append(model)
+
+    vae_values = set(_combo_values(info, "VAELoader", "vae_name"))
+    for model in REQUIRED_COMFY["vae"]:
+        if vae_values and model not in vae_values:
+            missing["vae"].append(model)
+
+    clip_values = set(_combo_values(info, "CLIPLoader", "clip_name"))
+    for model in REQUIRED_COMFY["clip"]:
+        if clip_values and model not in clip_values:
+            missing["clip"].append(model)
+
+    gguf_values = set(_combo_values(info, "CLIPLoaderGGUF", "clip_name"))
+    for model in REQUIRED_COMFY["clip_gguf"]:
+        if "CLIPLoaderGGUF" not in available_nodes:
+            missing["clip_gguf"].append(model)
+        elif gguf_values and model not in gguf_values:
+            missing["clip_gguf"].append(model)
+
+    sampler_values = set(_combo_values(info, "KSamplerSelect", "sampler_name"))
+    for sampler in REQUIRED_COMFY["samplers"]:
+        if sampler_values and sampler not in sampler_values:
+            missing["samplers"].append(sampler)
+
+    return missing
+
+
+def _input_name_by_slot(node, slot):
+    inputs = node.get("inputs", [])
+    if 0 <= slot < len(inputs):
+        return inputs[slot].get("name")
+    return None
+
+
+WIDGET_SCALAR_TYPES = {"INT", "FLOAT", "STRING", "BOOLEAN", "COMBO"}
+
+
+def _is_widget_type(type_spec):
+    # Combos arrive either as a raw list of options or as the literal "COMBO" string,
+    # depending on the ComfyUI version; both render a widget.
+    return isinstance(type_spec, list) or type_spec in WIDGET_SCALAR_TYPES
+
+
+def _make_api_entry(node, object_info):
+    """Convert a UI node into an API entry, mapping widget values to input names.
+
+    When ``object_info`` describes the node class, widget values are mapped to the
+    authoritative widget input order (combos and INT/FLOAT/STRING/BOOLEAN), skipping the
+    extra value that ``control_after_generate`` inputs (e.g. seeds) store. Links override
+    these defaults later. Falls back to the UI ``inputs`` widget flags when the class is
+    unknown.
+    """
+    entry = {"class_type": node["type"], "inputs": {}}
+    widgets = node.get("widgets_values", []) or []
+    if not isinstance(widgets, list):
+        return entry
+    spec = (object_info.get(node["type"], {}) or {}).get("input", {}) if object_info else {}
+    if spec:
+        ordered = list(spec.get("required", {}).items()) + list(spec.get("optional", {}).items())
+        idx = 0
+        for name, definition in ordered:
+            type_spec = definition[0] if definition else None
+            if not _is_widget_type(type_spec):
+                continue
+            if idx >= len(widgets):
+                break
+            entry["inputs"][name] = widgets[idx]
+            idx += 1
+            options = definition[1] if len(definition) > 1 and isinstance(definition[1], dict) else {}
+            if options.get("control_after_generate"):
+                idx += 1  # widgets_values stores the control value right after the widget
+        return entry
+    wi = 0
+    for inp in node.get("inputs", []):
+        if inp.get("widget"):
+            if wi < len(widgets):
+                entry["inputs"][inp["name"]] = widgets[wi]
+            wi += 1
+    return entry
+
+
+def workflow_to_api_prompt(workflow, compact_caption, seed, object_info=None):
+    """Convert the bundled UI workflow (with its subgraph) into a ComfyUI API prompt.
+
+    Subgraph instances are flattened: internal nodes are namespaced, internal links are
+    wired by id, and the subgraph boundary (instance inputs/outputs) is resolved so that
+    top-level wiring crosses into the subgraph correctly. The builder's caption is injected
+    into CLIPTextEncode (which prunes the original prompt-builder branch) and a fresh seed
+    into RandomNoise.
+    """
+    subgraph_defs = {sg["id"]: sg for sg in workflow.get("definitions", {}).get("subgraphs", [])}
+    prompt = {}
+
+    def key(scope, nid):
+        return str(nid) if scope is None else f"{scope}_{nid}"
+
+    instances = []  # (instance_node, subgraph_def, scope)
+    for node in workflow.get("nodes", []):
+        node_type = node.get("type")
+        if node_type == "MarkdownNote":
+            continue
+        if node_type in subgraph_defs:
+            sg = subgraph_defs[node_type]
+            scope = str(node["id"])
+            instances.append((node, sg, scope))
+            for child in sg.get("nodes", []):
+                if child.get("type") == "MarkdownNote":
+                    continue
+                prompt[key(scope, child["id"])] = _make_api_entry(child, object_info)
+        else:
+            prompt[str(node["id"])] = _make_api_entry(node, object_info)
+
+    # Wire internal subgraph links and build boundary maps per instance.
+    boundary_in = {}   # scope -> {input_name: [(internal_node_id, internal_slot), ...]}
+    boundary_out = {}  # scope -> {output_name: (internal_node_id, internal_slot)}
+    for node, sg, scope in instances:
+        internal_ids = {n["id"] for n in sg.get("nodes", [])}
+        node_by_id = {n["id"]: n for n in sg.get("nodes", [])}
+        link_by_id = {l["id"]: l for l in sg.get("links", [])}
+
+        for link in sg.get("links", []):
+            origin, target = link.get("origin_id"), link.get("target_id")
+            if origin in internal_ids and target in internal_ids:
+                tnode = node_by_id.get(target)
+                name = _input_name_by_slot(tnode, link.get("target_slot", 0))
+                if name:
+                    prompt[key(scope, target)]["inputs"][name] = [key(scope, origin), link.get("origin_slot", 0)]
+
+        in_map = {}
+        for sg_input in sg.get("inputs", []):
+            targets = []
+            for lid in sg_input.get("linkIds", []):
+                link = link_by_id.get(lid)
+                if link and link.get("target_id") in internal_ids:
+                    targets.append((link["target_id"], link.get("target_slot", 0)))
+            in_map[sg_input["name"]] = targets
+        boundary_in[scope] = in_map
+
+        out_map = {}
+        for sg_output in sg.get("outputs", []):
+            for lid in sg_output.get("linkIds", []):
+                link = link_by_id.get(lid)
+                if link and link.get("origin_id") in internal_ids:
+                    out_map[sg_output["name"]] = (link["origin_id"], link.get("origin_slot", 0))
+                    break
+        boundary_out[scope] = out_map
+
+    instance_by_id = {str(node["id"]): (node, sg, scope) for node, sg, scope in instances}
+
+    def resolve_source(origin_id, origin_slot):
+        """Return [node_key, slot] for a link origin, crossing subgraph output boundaries."""
+        sid = str(origin_id)
+        if sid in instance_by_id:
+            node, sg, scope = instance_by_id[sid]
+            out_name = _input_name_by_slot({"inputs": node.get("outputs", [])}, origin_slot)
+            internal = boundary_out.get(scope, {}).get(out_name)
+            if internal:
+                return [key(scope, internal[0]), internal[1]]
+            return None
+        return [sid, origin_slot]
+
+    # Wire top-level links, crossing into subgraph instances where needed.
+    for link in workflow.get("links", []):
+        if not isinstance(link, list) or len(link) < 6:
+            continue
+        _lid, oid, oslot, tid, tslot, _type = link[:6]
+        source = resolve_source(oid, oslot)
+        if source is None:
+            continue
+        tkey = str(tid)
+        if tkey in instance_by_id:
+            node, sg, scope = instance_by_id[tkey]
+            inst_input = node.get("inputs", [])
+            in_name = inst_input[tslot].get("name") if 0 <= tslot < len(inst_input) else None
+            for (tnode, tnslot) in boundary_in.get(scope, {}).get(in_name, []):
+                name = _input_name_by_slot({"inputs": sg_node_inputs(sg, tnode)}, tnslot)
+                if name:
+                    prompt[key(scope, tnode)]["inputs"][name] = source
+        elif tkey in prompt:
+            tnode = next((n for n in workflow.get("nodes", []) if str(n.get("id")) == tkey), None)
+            name = _input_name_by_slot(tnode, tslot) if tnode else None
+            if name:
+                prompt[tkey]["inputs"][name] = source
+
+    # Inject builder data; overriding CLIPTextEncode.text prunes the prompt-builder branch.
+    for entry in prompt.values():
+        if entry["class_type"] == "CLIPTextEncode":
+            entry["inputs"]["text"] = compact_caption
+        elif entry["class_type"] == "RandomNoise":
+            entry["inputs"]["noise_seed"] = seed
+    return prompt
+
+
+def sg_node_inputs(sg, node_id):
+    for node in sg.get("nodes", []):
+        if node.get("id") == node_id:
+            return node.get("inputs", [])
+    return []
+
+
+def find_save_image_node(prompt):
+    for node_id, entry in prompt.items():
+        if entry.get("class_type") in ("SaveImage", "PreviewImage"):
+            return node_id
+    return None
+
+
+def comfy_generate(settings, workflow, compact_caption, seed, should_cancel=None):
+    """Submit the workflow to ComfyUI and return raw PNG bytes of the first output image."""
+    try:
+        object_info = comfy_object_info(settings)
+    except ComfyError:
+        object_info = None
+    prompt = workflow_to_api_prompt(workflow, compact_caption, seed, object_info)
+    try:
+        result = comfy_post(settings, "/prompt", {"prompt": prompt})
+    except urllib.error.HTTPError as error:
+        detail = ""
+        try:
+            detail = error.read().decode("utf-8", "replace")
+        except OSError:
+            pass
+        raise ComfyError(f"HTTP {error.code}: {detail[:400]}")
+    except (urllib.error.URLError, TimeoutError, OSError, json.JSONDecodeError) as error:
+        raise ComfyError(str(error))
+    prompt_id = result.get("prompt_id")
+    if not prompt_id:
+        raise ComfyError(json.dumps(result)[:300])
+
+    for _ in range(600):  # up to ~5 minutes
+        if should_cancel and should_cancel():
+            raise ComfyError("cancelled")
+        try:
+            history = comfy_get(settings, f"/history/{prompt_id}", timeout=10)
+        except (urllib.error.URLError, TimeoutError, OSError, json.JSONDecodeError) as error:
+            raise ComfyError(str(error))
+        if prompt_id in history:
+            outputs = history[prompt_id].get("outputs", {})
+            for node_output in outputs.values():
+                for image in node_output.get("images", []):
+                    query = urllib.parse.urlencode(
+                        {"filename": image["filename"], "subfolder": image.get("subfolder", ""),
+                         "type": image.get("type", "output")}
+                    )
+                    url = f"{comfy_base_url(settings)}/view?{query}"
+                    request = urllib.request.Request(url, headers={"User-Agent": "IdeogramPromptBuilder"})
+                    with urllib.request.urlopen(request, timeout=30) as response:
+                        return response.read()
+            raise ComfyError("No image in workflow output.")
+        time.sleep(0.5)
+    raise ComfyError("Timed out waiting for generation.")
+
+
+class GenerationThread(QThread):
+    finished_ok = pyqtSignal(bytes)
+    failed = pyqtSignal(str)
+
+    def __init__(self, settings, workflow, caption, seed, parent=None):
+        super().__init__(parent)
+        self.settings = settings
+        self.workflow = workflow
+        self.caption = caption
+        self.seed = seed
+        self._cancel = False
+
+    def cancel(self):
+        self._cancel = True
+
+    def run(self):
+        try:
+            data = comfy_generate(
+                self.settings, self.workflow, self.caption, self.seed, lambda: self._cancel,
+            )
+            self.finished_ok.emit(data)
+        except ComfyError as error:
+            self.failed.emit(str(error))
+        except Exception as error:  # noqa: BLE001 - surface anything to the UI
+            self.failed.emit(str(error))
+
+
+def load_library():
+    """Read the prompt library from disk, returning a list of entries."""
+    if LIBRARY_FILE.exists():
+        try:
+            with open(LIBRARY_FILE, "r", encoding="utf-8") as handle:
+                data = json.load(handle)
+            if isinstance(data, list):
+                return data
+        except (OSError, json.JSONDecodeError):
+            return []
+    return []
+
+
+def save_library(entries):
+    """Persist the prompt library to disk."""
+    with open(LIBRARY_FILE, "w", encoding="utf-8") as handle:
+        json.dump(entries, handle, ensure_ascii=False, indent=2)
+
+
+def preview_file(entry):
+    """Resolve an entry's preview image to an existing path, or None."""
+    name = entry.get("preview")
+    if not name:
+        return None
+    path = PREVIEW_DIR / name
+    return path if path.exists() else None
+
+
+def remove_preview_file(entry):
+    """Delete the preview image associated with an entry, if any."""
+    path = preview_file(entry)
+    if path:
+        try:
+            path.unlink()
+        except OSError:
+            pass
+    entry["preview"] = None
+
+
+def attach_preview(entry, parent):
+    """Pick an image file and copy it into PREVIEW_DIR as this entry's preview."""
+    path, _filter = QFileDialog.getOpenFileName(
+        parent,
+        tr("prev.pick_title"),
+        "",
+        tr("prev.filter"),
+    )
+    if not path:
+        return False
+    try:
+        PREVIEW_DIR.mkdir(parents=True, exist_ok=True)
+        remove_preview_file(entry)
+        target_name = f"{entry['id']}{Path(path).suffix.lower() or '.png'}"
+        shutil.copyfile(path, PREVIEW_DIR / target_name)
+    except OSError as error:
+        QMessageBox.warning(parent, tr("prev.title"), tr("prev.save_fail").format(err=error))
+        return False
+    entry["preview"] = target_name
+    entry["updated"] = datetime.now().isoformat(timespec="seconds")
+    return True
+
+
+class PaletteEditor(QWidget):
+    changed = pyqtSignal()
+
+    def __init__(self, limit, parent=None):
+        super().__init__(parent)
+        self.limit = limit
+        self._colors = []
+        self._syncing = False
+
+        layout = QVBoxLayout(self)
+        layout.setContentsMargins(0, 0, 0, 0)
+        layout.setSpacing(8)
+
+        top = QHBoxLayout()
+        self.line_edit = QLineEdit()
+        self.line_edit.setPlaceholderText(tr("pal.placeholder"))
+        self.add_button = QPushButton(tr("pal.add"))
+        self.add_button.clicked.connect(self.add_color)
+        self.line_edit.textChanged.connect(self._line_changed)
+        top.addWidget(self.line_edit, 1)
+        top.addWidget(self.add_button)
+        layout.addLayout(top)
+
+        bottom = QHBoxLayout()
+        self.swatch_row = QHBoxLayout()
+        self.swatch_row.setSpacing(6)
+        self.swatch_row.addStretch()
+        bottom.addLayout(self.swatch_row, 1)
+        self.counter_label = QLabel("")
+        self.counter_label.setStyleSheet("color:#7A847C;background:transparent;")
+        bottom.addWidget(self.counter_label)
+        layout.addLayout(bottom)
+        self._update_counter()
+
+    def text(self):
+        return palette_text(self._colors)
+
+    def colors(self):
+        return list(self._colors)
+
+    def set_text(self, text):
+        self.set_colors(parse_palette(text, self.limit))
+
+    def set_colors(self, colors):
+        self._colors = [normalize_hex(color) for color in colors if normalize_hex(color)][: self.limit]
+        self._sync_line()
+        self._render_swatches()
+        self.changed.emit()
+
+    def add_color(self):
+        initial = QColor(self._colors[-1] if self._colors else "#FFFFFF")
+        color = QColorDialog.getColor(initial, self, tr("pal.configure"))
+        if not color.isValid():
+            return
+        value = color.name().upper()
+        if value not in self._colors and len(self._colors) < self.limit:
+            self._colors.append(value)
+            self._sync_line()
+            self._render_swatches()
+            self.changed.emit()
+
+    def edit_color(self, index):
+        color = QColorDialog.getColor(QColor(self._colors[index]), self, tr("pal.configure"))
+        if not color.isValid():
+            return
+        self._colors[index] = color.name().upper()
+        self._sync_line()
+        self._render_swatches()
+        self.changed.emit()
+
+    def remove_color(self, index):
+        del self._colors[index]
+        self._sync_line()
+        self._render_swatches()
+        self.changed.emit()
+
+    def _line_changed(self):
+        if self._syncing:
+            return
+        self._colors = parse_palette(self.line_edit.text(), self.limit)
+        self._render_swatches()
+        self.changed.emit()
+
+    def _sync_line(self):
+        self._syncing = True
+        self.line_edit.setText(palette_text(self._colors))
+        self._syncing = False
+
+    def _update_counter(self):
+        count = len(self._colors)
+        self.counter_label.setText(tr("counter.colors").format(count=count, limit=self.limit))
+        # Highlight when any color is not a valid uppercase #RRGGBB or the limit is exceeded (item 10).
+        invalid = count > self.limit or any(not HEX_RE.match(c) for c in self._colors)
+        self.line_edit.setProperty("invalid", "true" if invalid else "false")
+        self.line_edit.style().unpolish(self.line_edit)
+        self.line_edit.style().polish(self.line_edit)
+
+    def _render_swatches(self):
+        self._update_counter()
+        while self.swatch_row.count() > 1:
+            item = self.swatch_row.takeAt(0)
+            if item.widget():
+                item.widget().deleteLater()
+
+        for index, color in enumerate(self._colors):
+            holder = QWidget()
+            holder.setObjectName("SwatchHolder")
+            row = QHBoxLayout(holder)
+            row.setContentsMargins(0, 0, 0, 0)
+            row.setSpacing(2)
+
+            swatch = QToolButton()
+            swatch.setToolTip(tr("pal.swatch_tip").format(color=color))
+            swatch.setFixedSize(34, 28)
+            swatch.setStyleSheet(
+                f"QToolButton {{ background: {color}; border: 1px solid #AEB8B1; border-radius: 6px; }}"
+            )
+            swatch.clicked.connect(lambda _checked=False, i=index: self.edit_color(i))
+            remove = QToolButton()
+            remove.setText("×")
+            remove.setToolTip(tr("pal.remove"))
+            remove.setFixedSize(22, 28)
+            remove.clicked.connect(lambda _checked=False, i=index: self.remove_color(i))
+            row.addWidget(swatch)
+            row.addWidget(remove)
+            self.swatch_row.insertWidget(index, holder)
+
+
+class BBoxCanvas(QFrame):
+    selected = pyqtSignal(int)
+    bbox_changed = pyqtSignal(int, list)
+
+    BASE_SIZE = 340
+
+    def __init__(self):
+        super().__init__()
+        self.elements = []
+        self.selected_index = None
+        self.drag_mode = None
+        self.drag_index = None
+        self.drag_start = QPointF()
+        self.start_bbox = None
+        self.zoom = 1.0
+        self.ref_pixmap = None
+        self.theme = THEMES["light"]
+        self.setMinimumHeight(360)
+        self.setMouseTracking(True)
+        self.setCursor(Qt.CursorShape.CrossCursor)
+
+    def set_data(self, elements, selected_index):
+        self.elements = elements
+        self.selected_index = selected_index
+        self.update()
+
+    def set_theme(self, theme):
+        self.theme = THEMES.get(theme, THEMES["light"])
+        self.update()
+
+    def set_reference(self, pixmap):
+        """Set (or clear with None) the background reference image; scales with the grid."""
+        self.ref_pixmap = pixmap if pixmap and not pixmap.isNull() else None
+        self.update()
+
+    def set_zoom(self, percent):
+        self.zoom = max(0.5, min(3.0, percent / 100.0))
+        size = int(self.BASE_SIZE * self.zoom)
+        margin = 16
+        # Grow minimums so the surrounding scroll area exposes scrollbars when zoomed in.
+        self.setMinimumHeight(max(360, size + margin * 2))
+        self.setMinimumWidth(size + margin * 2 if self.zoom > 1.0 else 0)
+        self.update()
+
+    def canvas_rect(self):
+        margin = 16
+        size = self.BASE_SIZE * self.zoom
+        left = max(margin, (self.width() - size) / 2)
+        top = margin
+        return QRectF(left, top, size, size)
+
+    def bbox_to_rect(self, bbox):
+        canvas = self.canvas_rect()
+        y1, x1, y2, x2 = bbox
+        return QRectF(
+            canvas.left() + canvas.width() * x1 / 1000,
+            canvas.top() + canvas.height() * y1 / 1000,
+            canvas.width() * (x2 - x1) / 1000,
+            canvas.height() * (y2 - y1) / 1000,
+        )
+
+    def point_to_bbox_delta(self, delta):
+        canvas = self.canvas_rect()
+        return delta.y() * 1000 / canvas.height(), delta.x() * 1000 / canvas.width()
+
+    def hit_handle(self, point, rect):
+        handles = {
+            "nw": rect.topLeft(),
+            "n": QPointF(rect.center().x(), rect.top()),
+            "ne": rect.topRight(),
+            "e": QPointF(rect.right(), rect.center().y()),
+            "se": rect.bottomRight(),
+            "s": QPointF(rect.center().x(), rect.bottom()),
+            "sw": rect.bottomLeft(),
+            "w": QPointF(rect.left(), rect.center().y()),
+        }
+        for name, handle in handles.items():
+            if QRectF(handle.x() - 7, handle.y() - 7, 14, 14).contains(point):
+                return name
+        return None
+
+    def hit_test(self, point):
+        for index in range(len(self.elements) - 1, -1, -1):
+            element = self.elements[index]
+            if not element.get("use_bbox"):
+                continue
+            rect = self.bbox_to_rect(element["bbox"])
+            handle = self.hit_handle(point, rect)
+            if handle:
+                return index, handle
+            if rect.contains(point):
+                return index, "move"
+        return None, None
+
+    def mousePressEvent(self, event):
+        if event.button() != Qt.MouseButton.LeftButton:
+            return
+        index, mode = self.hit_test(event.position())
+        if index is None:
+            return
+        self.drag_index = index
+        self.drag_mode = mode
+        self.drag_start = event.position()
+        self.start_bbox = list(self.elements[index]["bbox"])
+        self.selected.emit(index)
+
+    def mouseMoveEvent(self, event):
+        if self.drag_index is None:
+            index, mode = self.hit_test(event.position())
+            self.setCursor(self.cursor_for_mode(mode))
+            return
+
+        dy, dx = self.point_to_bbox_delta(event.position() - self.drag_start)
+        y1, x1, y2, x2 = self.start_bbox
+
+        if self.drag_mode == "move":
+            height = y2 - y1
+            width = x2 - x1
+            y1 = clamp(y1 + dy, 0, 1000 - height)
+            x1 = clamp(x1 + dx, 0, 1000 - width)
+            y2 = y1 + height
+            x2 = x1 + width
+        else:
+            if "n" in self.drag_mode:
+                y1 = clamp(y1 + dy, 0, y2 - MIN_BBOX_SIZE)
+            if "s" in self.drag_mode:
+                y2 = clamp(y2 + dy, y1 + MIN_BBOX_SIZE, 1000)
+            if "w" in self.drag_mode:
+                x1 = clamp(x1 + dx, 0, x2 - MIN_BBOX_SIZE)
+            if "e" in self.drag_mode:
+                x2 = clamp(x2 + dx, x1 + MIN_BBOX_SIZE, 1000)
+
+        self.bbox_changed.emit(self.drag_index, [y1, x1, y2, x2])
+
+    def mouseReleaseEvent(self, event):
+        self.drag_index = None
+        self.drag_mode = None
+        self.start_bbox = None
+        self.setCursor(Qt.CursorShape.CrossCursor)
+
+    def cursor_for_mode(self, mode):
+        mapping = {
+            "move": Qt.CursorShape.SizeAllCursor,
+            "n": Qt.CursorShape.SizeVerCursor,
+            "s": Qt.CursorShape.SizeVerCursor,
+            "e": Qt.CursorShape.SizeHorCursor,
+            "w": Qt.CursorShape.SizeHorCursor,
+            "nw": Qt.CursorShape.SizeFDiagCursor,
+            "se": Qt.CursorShape.SizeFDiagCursor,
+            "ne": Qt.CursorShape.SizeBDiagCursor,
+            "sw": Qt.CursorShape.SizeBDiagCursor,
+        }
+        return mapping.get(mode, Qt.CursorShape.CrossCursor)
+
+    def paintEvent(self, event):
+        super().paintEvent(event)
+        painter = QPainter(self)
+        painter.setRenderHint(QPainter.RenderHint.Antialiasing)
+        canvas = self.canvas_rect()
+
+        painter.setPen(QPen(QColor(self.theme["canvas_grid"]), 1))
+        painter.setBrush(QColor(self.theme["canvas_bg"]))
+        painter.drawRoundedRect(canvas, 10, 10)
+
+        # Reference image fills the grid square and therefore scales with the zoom.
+        if self.ref_pixmap is not None:
+            scaled = self.ref_pixmap.scaled(
+                int(canvas.width()), int(canvas.height()),
+                Qt.AspectRatioMode.KeepAspectRatio,
+                Qt.TransformationMode.SmoothTransformation,
+            )
+            img_x = canvas.left() + (canvas.width() - scaled.width()) / 2
+            img_y = canvas.top() + (canvas.height() - scaled.height()) / 2
+            painter.setOpacity(0.85)
+            painter.drawPixmap(int(img_x), int(img_y), scaled)
+            painter.setOpacity(1.0)
+
+        painter.setPen(QPen(QColor(self.theme["canvas_grid"]), 1))
+        for step in range(1, 10):
+            x = canvas.left() + canvas.width() * step / 10
+            y = canvas.top() + canvas.height() * step / 10
+            painter.drawLine(int(x), int(canvas.top()), int(x), int(canvas.bottom()))
+            painter.drawLine(int(canvas.left()), int(y), int(canvas.right()), int(y))
+
+        painter.setPen(QPen(QColor(self.theme["canvas_label"]), 1))
+        painter.drawText(int(canvas.left()) + 10, int(canvas.top()) + 22, tr("canvas.label"))
+
+        for index, element in enumerate(self.elements):
+            if not element.get("use_bbox"):
+                continue
+            rect = self.bbox_to_rect(element["bbox"])
+            base = QColor("#C470A8") if element["type"] == "text" else QColor(self.theme["accent"])
+            fill = QColor(base)
+            fill.setAlpha(32)
+            painter.setBrush(fill)
+            painter.setPen(QPen(base, 3 if index == self.selected_index else 2))
+            painter.drawRoundedRect(rect, 6, 6)
+            painter.setPen(base)
+            painter.drawText(rect.adjusted(7, 5, -7, -5), Qt.AlignmentFlag.AlignLeft, element.get("label") or str(index + 1))
+
+            if index == self.selected_index:
+                painter.setBrush(QColor(self.theme["panel"]))
+                painter.setPen(QPen(base, 2))
+                for point in [
+                    rect.topLeft(),
+                    QPointF(rect.center().x(), rect.top()),
+                    rect.topRight(),
+                    QPointF(rect.right(), rect.center().y()),
+                    rect.bottomRight(),
+                    QPointF(rect.center().x(), rect.bottom()),
+                    rect.bottomLeft(),
+                    QPointF(rect.left(), rect.center().y()),
+                ]:
+                    painter.drawEllipse(point, 5, 5)
+
+
+class ComfySettingsDialog(QDialog):
+    """Edit ComfyUI connection settings, persisted to comfy_settings.json."""
+
+    def __init__(self, settings, parent=None):
+        super().__init__(parent)
+        self.settings = dict(settings)
+        self.setWindowTitle(tr("set.title"))
+        self.setMinimumWidth(420)
+
+        layout = QVBoxLayout(self)
+        form = QFormLayout()
+        self.host_edit = QLineEdit(str(self.settings.get("comfy_host", "127.0.0.1")))
+        self.port_spin = QSpinBox()
+        self.port_spin.setRange(1, 65535)
+        self.port_spin.setValue(int(self.settings.get("comfy_port", 8188)))
+        self.https_check = QCheckBox(tr("set.https"))
+        self.https_check.setChecked(bool(self.settings.get("comfy_https", False)))
+        form.addRow(tr("set.host"), self.host_edit)
+        form.addRow(tr("set.port"), self.port_spin)
+        form.addRow("", self.https_check)
+        layout.addLayout(form)
+
+        test_row = QHBoxLayout()
+        self.test_button = QPushButton(tr("set.test"))
+        self.test_button.clicked.connect(self.test_connection)
+        test_row.addWidget(self.test_button)
+        test_row.addStretch()
+        layout.addLayout(test_row)
+
+        buttons = QDialogButtonBox(
+            QDialogButtonBox.StandardButton.Save | QDialogButtonBox.StandardButton.Cancel
+        )
+        buttons.accepted.connect(self.accept)
+        buttons.rejected.connect(self.reject)
+        layout.addWidget(buttons)
+
+    def values(self):
+        return {
+            "comfy_host": self.host_edit.text().strip() or "127.0.0.1",
+            "comfy_port": self.port_spin.value(),
+            "comfy_https": self.https_check.isChecked(),
+        }
+
+    def test_connection(self):
+        probe = dict(self.settings)
+        probe.update(self.values())
+        try:
+            comfy_test_connection(probe)
+        except ComfyError as error:
+            QMessageBox.warning(self, tr("set.title"), tr("set.test_fail").format(err=error))
+            return
+        QMessageBox.information(self, tr("set.title"), tr("set.test_ok"))
+
+
+class LibraryDialog(QDialog):
+    """Browse the prompt library: load, rename, attach a preview, or delete entries."""
+
+    def __init__(self, entries, parent=None):
+        super().__init__(parent)
+        self.entries = entries
+        self.selected_caption = None
+        self.selected_id = None
+        self._filtered = []  # list of original indices currently shown
+        self.setWindowTitle(tr("libd.title"))
+        self.resize(900, 600)
+
+        layout = QHBoxLayout(self)
+        layout.setSpacing(14)
+
+        left = QVBoxLayout()
+        left.addWidget(QLabel(tr("libd.saved_prompts")))
+        self.search_edit = QLineEdit()
+        self.search_edit.setPlaceholderText(tr("libd.search"))
+        self.search_edit.textChanged.connect(lambda _t: self._refresh_list(0))
+        left.addWidget(self.search_edit)
+        self.list_widget = QListWidget()
+        self.list_widget.setMinimumWidth(300)
+        self.list_widget.currentRowChanged.connect(self._show_details)
+        self.list_widget.itemDoubleClicked.connect(lambda _item: self.use_selected())
+        left.addWidget(self.list_widget, 1)
+        io_row = QHBoxLayout()
+        export_button = QPushButton(tr("libd.export"))
+        export_button.clicked.connect(self.export_library)
+        import_button = QPushButton(tr("libd.import"))
+        import_button.clicked.connect(self.import_library)
+        io_row.addWidget(export_button)
+        io_row.addWidget(import_button)
+        left.addLayout(io_row)
+        layout.addLayout(left, 1)
+
+        right = QVBoxLayout()
+        right.setSpacing(10)
+        self.preview_label = QLabel(tr("libd.no_preview"))
+        self.preview_label.setAlignment(Qt.AlignmentFlag.AlignCenter)
+        self.preview_label.setMinimumSize(360, 260)
+        self.preview_label.setStyleSheet(
+            "background:palette(base);border:1px solid palette(mid);border-radius:8px;"
+        )
+        right.addWidget(self.preview_label, 1)
+
+        self.meta_label = QLabel("")
+        self.meta_label.setWordWrap(True)
+        right.addWidget(self.meta_label)
+
+        right.addWidget(QLabel(tr("libd.tags")))
+        self.tags_edit = QLineEdit()
+        self.tags_edit.editingFinished.connect(self._save_tags)
+        right.addWidget(self.tags_edit)
+
+        button_row = QHBoxLayout()
+        self.use_button = QPushButton(tr("libd.use"))
+        self.use_button.setObjectName("PrimaryButton")
+        self.use_button.clicked.connect(self.use_selected)
+        self.rename_button = QPushButton(tr("libd.rename"))
+        self.rename_button.clicked.connect(self.rename_selected)
+        button_row.addWidget(self.use_button)
+        button_row.addWidget(self.rename_button)
+        right.addLayout(button_row)
+
+        button_row2 = QHBoxLayout()
+        self.preview_button = QPushButton(tr("libd.set_preview"))
+        self.preview_button.clicked.connect(self.set_preview)
+        self.paste_preview_button = QPushButton(tr("libd.paste_preview"))
+        self.paste_preview_button.clicked.connect(self.paste_preview)
+        self.clear_preview_button = QPushButton(tr("libd.clear_preview"))
+        self.clear_preview_button.clicked.connect(self.clear_preview)
+        button_row2.addWidget(self.preview_button)
+        button_row2.addWidget(self.paste_preview_button)
+        button_row2.addWidget(self.clear_preview_button)
+        right.addLayout(button_row2)
+
+        button_row3 = QHBoxLayout()
+        self.delete_button = QPushButton(tr("libd.delete"))
+        self.delete_button.clicked.connect(self.delete_selected)
+        close_button = QPushButton(tr("libd.close"))
+        close_button.clicked.connect(self.reject)
+        button_row3.addWidget(self.delete_button)
+        button_row3.addStretch()
+        button_row3.addWidget(close_button)
+        right.addLayout(button_row3)
+        layout.addLayout(right, 1)
+
+        self._refresh_list(0 if self.entries else -1)
+
+    def _matches(self, entry, query):
+        if not query:
+            return True
+        haystack = " ".join([
+            entry.get("name", ""),
+            " ".join(entry.get("tags", []) or []),
+            entry.get("caption", {}).get("high_level_description", ""),
+        ]).lower()
+        return query in haystack
+
+    def _refresh_list(self, select_row):
+        query = self.search_edit.text().strip().lower()
+        self.list_widget.blockSignals(True)
+        self.list_widget.clear()
+        self._filtered = []
+        for index, entry in enumerate(self.entries):
+            if not self._matches(entry, query):
+                continue
+            self._filtered.append(index)
+            mark = "🖼 " if preview_file(entry) else ""
+            tags = entry.get("tags", []) or []
+            suffix = f"  [{', '.join(tags)}]" if tags else ""
+            self.list_widget.addItem(QListWidgetItem(f"{mark}{entry.get('name') or tr('lib.untitled')}{suffix}"))
+        self.list_widget.blockSignals(False)
+        if 0 <= select_row < len(self._filtered):
+            self.list_widget.setCurrentRow(select_row)
+        else:
+            self._show_details(self.list_widget.currentRow())
+
+    def _current_entry(self):
+        row = self.list_widget.currentRow()
+        if 0 <= row < len(self._filtered):
+            original = self._filtered[row]
+            return original, self.entries[original]
+        return None, None
+
+    def _show_details(self, row):
+        has = 0 <= row < len(self._filtered)
+        for button in (self.use_button, self.rename_button, self.preview_button,
+                       self.paste_preview_button, self.clear_preview_button, self.delete_button):
+            button.setEnabled(has)
+        self.tags_edit.setEnabled(has)
+        if not has:
+            self.preview_label.setText(tr("libd.no_preview"))
+            self.preview_label.setPixmap(QPixmap())
+            self.meta_label.setText("")
+            self.tags_edit.blockSignals(True)
+            self.tags_edit.clear()
+            self.tags_edit.blockSignals(False)
+            return
+        entry = self.entries[self._filtered[row]]
+        self.tags_edit.blockSignals(True)
+        self.tags_edit.setText(", ".join(entry.get("tags", []) or []))
+        self.tags_edit.blockSignals(False)
+        path = preview_file(entry)
+        if path:
+            pixmap = QPixmap(str(path))
+            if not pixmap.isNull():
+                self.preview_label.setPixmap(
+                    pixmap.scaled(
+                        self.preview_label.size(),
+                        Qt.AspectRatioMode.KeepAspectRatio,
+                        Qt.TransformationMode.SmoothTransformation,
+                    )
+                )
+            else:
+                self.preview_label.setText(tr("libd.preview_unavailable"))
+        else:
+            self.preview_label.setPixmap(QPixmap())
+            self.preview_label.setText(tr("libd.no_preview"))
+        caption = entry.get("caption", {})
+        high = caption.get("high_level_description", "") or tr("libd.no_high")
+        count = len(caption.get("compositional_deconstruction", {}).get("elements", []))
+        updated = entry.get("updated", entry.get("created", ""))
+        self.meta_label.setText(tr("libd.meta").format(updated=updated, count=count, high=high))
+
+    def use_selected(self):
+        _row, entry = self._current_entry()
+        if entry is None:
+            return
+        self.selected_caption = entry.get("caption", {})
+        self.selected_id = entry.get("id")
+        self.accept()
+
+    def rename_selected(self):
+        _row, entry = self._current_entry()
+        if entry is None:
+            return
+        name, ok = QInputDialog.getText(self, tr("libd.rename_title"), tr("libd.rename_label"), text=entry.get("name", ""))
+        if ok and name.strip():
+            entry["name"] = name.strip()
+            entry["updated"] = datetime.now().isoformat(timespec="seconds")
+            save_library(self.entries)
+            self._refresh_list(self.list_widget.currentRow())
+
+    def _save_tags(self):
+        _row, entry = self._current_entry()
+        if entry is None:
+            return
+        tags = [t.strip() for t in self.tags_edit.text().split(",") if t.strip()]
+        if tags != (entry.get("tags") or []):
+            entry["tags"] = tags
+            entry["updated"] = datetime.now().isoformat(timespec="seconds")
+            save_library(self.entries)
+            self._refresh_list(self.list_widget.currentRow())
+
+    def set_preview(self):
+        _row, entry = self._current_entry()
+        if entry is None:
+            return
+        if attach_preview(entry, self):
+            save_library(self.entries)
+            self._refresh_list(self.list_widget.currentRow())
+
+    def paste_preview(self):
+        _row, entry = self._current_entry()
+        if entry is None:
+            return
+        image = QGuiApplication.clipboard().image()
+        if image.isNull():
+            QMessageBox.information(self, tr("libd.paste_preview"), tr("libd.no_clipboard_image"))
+            return
+        try:
+            PREVIEW_DIR.mkdir(parents=True, exist_ok=True)
+            remove_preview_file(entry)
+            target_name = f"{entry['id']}.png"
+            image.save(str(PREVIEW_DIR / target_name), "PNG")
+        except OSError as error:
+            QMessageBox.warning(self, tr("prev.title"), tr("prev.save_fail").format(err=error))
+            return
+        entry["preview"] = target_name
+        entry["updated"] = datetime.now().isoformat(timespec="seconds")
+        save_library(self.entries)
+        self._refresh_list(self.list_widget.currentRow())
+
+    def clear_preview(self):
+        _row, entry = self._current_entry()
+        if entry is None or not entry.get("preview"):
+            return
+        remove_preview_file(entry)
+        entry["updated"] = datetime.now().isoformat(timespec="seconds")
+        save_library(self.entries)
+        self._refresh_list(self.list_widget.currentRow())
+
+    def delete_selected(self):
+        original, entry = self._current_entry()
+        if entry is None:
+            return
+        confirm = QMessageBox.question(
+            self,
+            tr("libd.delete_title"),
+            tr("libd.delete_q").format(name=entry.get("name", "")),
+        )
+        if confirm != QMessageBox.StandardButton.Yes:
+            return
+        row = self.list_widget.currentRow()
+        remove_preview_file(entry)
+        del self.entries[original]
+        save_library(self.entries)
+        self._refresh_list(min(row, len(self._filtered) - 1))
+
+    def export_library(self):
+        path, _filter = QFileDialog.getSaveFileName(
+            self, tr("libd.export"), "prompt_library.zip", tr("libd.export_filter")
+        )
+        if not path:
+            return
+        try:
+            with zipfile.ZipFile(path, "w", zipfile.ZIP_DEFLATED) as archive:
+                archive.writestr("prompt_library.json",
+                                 json.dumps(self.entries, ensure_ascii=False, indent=2))
+                for entry in self.entries:
+                    preview = preview_file(entry)
+                    if preview:
+                        archive.write(preview, f"prompt_previews/{preview.name}")
+        except OSError as error:
+            QMessageBox.warning(self, tr("libd.export"), tr("libd.export_fail").format(err=error))
+            return
+        QMessageBox.information(self, tr("libd.export"), tr("libd.export_done").format(path=path))
+
+    def import_library(self):
+        path, _filter = QFileDialog.getOpenFileName(
+            self, tr("libd.import"), "", tr("libd.export_filter")
+        )
+        if not path:
+            return
+        try:
+            with zipfile.ZipFile(path, "r") as archive:
+                imported = json.loads(archive.read("prompt_library.json").decode("utf-8"))
+                PREVIEW_DIR.mkdir(parents=True, exist_ok=True)
+                existing_ids = {e.get("id") for e in self.entries}
+                added = 0
+                for entry in imported:
+                    if not isinstance(entry, dict):
+                        continue
+                    if entry.get("id") in existing_ids:
+                        entry["id"] = uuid.uuid4().hex
+                    preview = entry.get("preview")
+                    if preview:
+                        member = f"prompt_previews/{preview}"
+                        if member in archive.namelist():
+                            with archive.open(member) as source:
+                                (PREVIEW_DIR / preview).write_bytes(source.read())
+                    self.entries.append(entry)
+                    existing_ids.add(entry.get("id"))
+                    added += 1
+            save_library(self.entries)
+        except (OSError, KeyError, json.JSONDecodeError, zipfile.BadZipFile) as error:
+            QMessageBox.warning(self, tr("libd.import"), tr("libd.import_fail").format(err=error))
+            return
+        self._refresh_list(0)
+        QMessageBox.information(self, tr("libd.import"), tr("libd.import_done").format(count=added))
+
+
+class PromptBuilder(QMainWindow):
+    def __init__(self):
+        super().__init__()
+        self.elements = []
+        self.selected_index = None
+        self._loading = False
+        self._toolbar = None
+        self.settings = load_settings()
+        self.theme = self.settings.get("theme", "light")
+        self._undo_stack = []
+        self._redo_stack = []
+        self._suspend_history = False
+        self._library_entry_id = None  # id of the entry loaded from the library (item 7)
+        self._gen_thread = None
+        self.setWindowTitle(tr("app.title"))
+        self.resize(1460, 900)
+        self._build_ui()
+        if not self._restore_draft():
+            self.load_caption(EXAMPLE_CAPTION)
+        self._push_history(initial=True)
+
+    def set_language(self, language):
+        global CURRENT_LANG
+        if language == CURRENT_LANG or language not in TRANSLATIONS:
+            return
+        caption = self.current_caption()
+        ref = self.canvas.ref_pixmap if hasattr(self, "canvas") else None
+        CURRENT_LANG = language
+        self.settings["language"] = language
+        save_settings(self.settings)
+        self.setWindowTitle(tr("app.title"))
+        if self._toolbar is not None:
+            self.removeToolBar(self._toolbar)
+            self._toolbar.deleteLater()
+            self._toolbar = None
+        self._suspend_history = True
+        self._build_ui()
+        self.load_caption(caption)
+        if ref is not None:
+            self.canvas.set_reference(ref)
+        self._suspend_history = False
+
+    def toggle_theme(self):
+        self.theme = "dark" if self.theme == "light" else "light"
+        self.settings["theme"] = self.theme
+        save_settings(self.settings)
+        self.setStyleSheet(build_stylesheet(self.theme))
+        self.canvas.set_theme(self.theme)
+
+    # --- Undo / redo (item 1) -------------------------------------------
+    def _snapshot(self):
+        return copy.deepcopy(self.current_caption())
+
+    def _push_history(self, initial=False):
+        if self._suspend_history:
+            return
+        snap = self._snapshot()
+        if self._undo_stack and self._undo_stack[-1] == snap:
+            return
+        self._undo_stack.append(snap)
+        if len(self._undo_stack) > MAX_UNDO:
+            self._undo_stack.pop(0)
+        if not initial:
+            self._redo_stack.clear()
+
+    def undo(self):
+        if len(self._undo_stack) < 2:
+            return
+        self._redo_stack.append(self._undo_stack.pop())
+        target = copy.deepcopy(self._undo_stack[-1])
+        self._suspend_history = True
+        self.load_caption(target)
+        self._suspend_history = False
+
+    def redo(self):
+        if not self._redo_stack:
+            return
+        target = self._redo_stack.pop()
+        self._undo_stack.append(copy.deepcopy(target))
+        self._suspend_history = True
+        self.load_caption(copy.deepcopy(target))
+        self._suspend_history = False
+
+    def install_translate_menu(self, widget):
+        widget.setContextMenuPolicy(Qt.ContextMenuPolicy.CustomContextMenu)
+        widget.customContextMenuRequested.connect(lambda point, target=widget: self.show_translate_menu(target, point))
+
+    def show_translate_menu(self, widget, point):
+        menu = widget.createStandardContextMenu()
+        selected = self.selected_text(widget)
+        if selected:
+            menu.addSeparator()
+            ru_action = menu.addAction(tr("trn.to_ru"))
+            en_action = menu.addAction(tr("trn.to_en"))
+            ru_action.triggered.connect(lambda: self.translate_selection(widget, "ru"))
+            en_action.triggered.connect(lambda: self.translate_selection(widget, "en"))
+        menu.exec(widget.mapToGlobal(point))
+
+    def selected_text(self, widget):
+        if isinstance(widget, QLineEdit):
+            return widget.selectedText()
+        if isinstance(widget, (QTextEdit, QPlainTextEdit)):
+            return widget.textCursor().selectedText().replace("\u2029", "\n")
+        return ""
+
+    def replace_selection(self, widget, replacement):
+        if isinstance(widget, QLineEdit):
+            widget.insert(replacement)
+            return
+        cursor = widget.textCursor()
+        cursor.insertText(replacement)
+        widget.setTextCursor(cursor)
+
+    def translate_selection(self, widget, target_language):
+        selected = self.selected_text(widget)
+        if not selected.strip():
+            return
+        try:
+            translated = google_translate_text(selected, target_language)
+        except (urllib.error.URLError, TimeoutError, json.JSONDecodeError, IndexError, KeyError, TypeError) as error:
+            QMessageBox.warning(self, tr("trn.error_title"), tr("trn.error_msg").format(err=error))
+            return
+        if translated:
+            self.replace_selection(widget, translated)
+
+    def _make_action(self, title, callback, shortcut=None):
+        action = QAction(title, self)
+        action.triggered.connect(callback)
+        if shortcut:
+            action.setShortcut(QKeySequence(shortcut))
+        return action
+
+    def _build_menubar(self):
+        bar = self.menuBar()
+        bar.clear()
+        file_menu = bar.addMenu(tr("menu.file"))
+        file_menu.addAction(self._make_action(tr("tb.example"),
+                            lambda: self.load_caption(EXAMPLE_CAPTION, mark_history=True)))
+        file_menu.addAction(self._make_action(tr("tb.import"), self.import_json))
+        file_menu.addAction(self._make_action(tr("tb.save_json"), self.save_json, "Ctrl+S"))
+        file_menu.addAction(self._make_action(tr("tb.copy"), self.copy_current_json))
+
+        edit_menu = bar.addMenu(tr("menu.edit"))
+        edit_menu.addAction(self._make_action(tr("tb.undo"), self.undo, "Ctrl+Z"))
+        edit_menu.addAction(self._make_action(tr("tb.redo"), self.redo, "Ctrl+Y"))
+
+        lib_menu = bar.addMenu(tr("menu.library"))
+        lib_menu.addAction(self._make_action(tr("tb.save_library"), self.save_to_library))
+        lib_menu.addAction(self._make_action(tr("tb.overwrite"), self.overwrite_in_library))
+        lib_menu.addAction(self._make_action(tr("tb.library"), self.open_library))
+
+        comfy_menu = bar.addMenu(tr("menu.comfy"))
+        comfy_menu.addAction(self._make_action(tr("tb.comfy_settings"), self.open_comfy_settings))
+        comfy_menu.addAction(self._make_action(tr("tb.check_comfy"), self.check_comfy))
+        comfy_menu.addAction(self._make_action(tr("tb.generate"), self.generate_in_comfy))
+
+        view_menu = bar.addMenu(tr("menu.view"))
+        view_menu.addAction(self._make_action(tr("tb.theme"), self.toggle_theme))
+
+    def _build_ui(self):
+        self.setStyleSheet(build_stylesheet(self.theme))
+        self._build_menubar()
+        toolbar = self.addToolBar("Main")
+        toolbar.setMovable(False)
+        self._toolbar = toolbar
+        # Slim toolbar: the most frequent actions only; everything lives in the menus too.
+        generate_action = self._make_action(tr("tb.generate"), self.generate_in_comfy)
+        toolbar.addAction(generate_action)
+        toolbar.addSeparator()
+        toolbar.addAction(self._make_action(tr("tb.undo"), self.undo))
+        toolbar.addAction(self._make_action(tr("tb.redo"), self.redo))
+        toolbar.addSeparator()
+        toolbar.addAction(self._make_action(tr("tb.save_library"), self.save_to_library))
+        toolbar.addAction(self._make_action(tr("tb.library"), self.open_library))
+        toolbar.addSeparator()
+        toolbar.addAction(self._make_action(tr("tb.copy"), self.copy_current_json))
+
+        spacer = QWidget()
+        spacer.setSizePolicy(QSizePolicy.Policy.Expanding, QSizePolicy.Policy.Preferred)
+        toolbar.addWidget(spacer)
+        toolbar.addWidget(QLabel(tr("tb.language") + " "))
+        self.language_combo = QComboBox()
+        for code in available_languages():
+            self.language_combo.addItem(LANGUAGE_NAMES.get(code, code), code)
+        current_index = self.language_combo.findData(CURRENT_LANG)
+        if current_index >= 0:
+            self.language_combo.setCurrentIndex(current_index)
+        self.language_combo.currentIndexChanged.connect(
+            lambda _i: self.set_language(self.language_combo.currentData())
+        )
+        toolbar.addWidget(self.language_combo)
+        theme_action = QAction(tr("tb.theme"), self)
+        theme_action.triggered.connect(self.toggle_theme)
+        toolbar.addAction(theme_action)
+
+        splitter = QSplitter(Qt.Orientation.Horizontal)
+        self.setCentralWidget(splitter)
+
+        editor_scroll = QScrollArea()
+        editor_scroll.setWidgetResizable(True)
+        editor_widget = QWidget()
+        self.editor_layout = QVBoxLayout(editor_widget)
+        self.editor_layout.setContentsMargins(16, 16, 16, 16)
+        self.editor_layout.setSpacing(12)
+        editor_scroll.setWidget(editor_widget)
+        splitter.addWidget(editor_scroll)
+
+        output_widget = QWidget()
+        output_layout = QVBoxLayout(output_widget)
+        output_layout.setContentsMargins(16, 16, 16, 16)
+        output_layout.setSpacing(10)
+        splitter.addWidget(output_widget)
+        splitter.setSizes([900, 560])
+
+        self._build_summary()
+        self._build_presets()
+        self._build_style()
+        self._build_composition()
+        self.editor_layout.addStretch()
+        self._build_output(output_layout)
+
+    def _build_summary(self):
+        box = QGroupBox(tr("grp.high"))
+        layout = QVBoxLayout(box)
+        self.high_text = QTextEdit()
+        self.high_text.setMinimumHeight(110)
+        self.high_text.setPlaceholderText(tr("high.placeholder"))
+        self.high_text.textChanged.connect(self.update_output)
+        self.install_translate_menu(self.high_text)
+        layout.addWidget(self.high_text)
+        self.editor_layout.addWidget(box)
+
+    def _build_presets(self):
+        box = QGroupBox(tr("grp.presets"))
+        layout = QGridLayout(box)
+        layout.setSpacing(8)
+        for index, name in enumerate(PROMPT_PRESETS):
+            button = QPushButton(name)
+            button.clicked.connect(lambda _checked=False, value=name: self.apply_preset(value))
+            layout.addWidget(button, index // 2, index % 2)
+        no_safety = QPushButton(tr("preset.no_safety"))
+        no_safety.clicked.connect(self.append_no_safety_filter)
+        layout.addWidget(no_safety, 2, 0, 1, 2)
+        self.editor_layout.addWidget(box)
+
+    def _build_style(self):
+        box = QGroupBox(tr("grp.style"))
+        layout = QVBoxLayout(box)
+        mode_row = QHBoxLayout()
+        self.photo_radio = QRadioButton(tr("style.photo"))
+        self.art_radio = QRadioButton(tr("style.art"))
+        self.photo_radio.setChecked(True)
+        self.photo_radio.toggled.connect(self._style_mode_changed)
+        mode_row.addWidget(self.photo_radio)
+        mode_row.addWidget(self.art_radio)
+        mode_row.addStretch()
+        layout.addLayout(mode_row)
+
+        form = QFormLayout()
+        self.aesthetics_edit = QLineEdit()
+        self.lighting_edit = QLineEdit()
+        self.photo_edit = QLineEdit()
+        self.art_style_edit = QLineEdit()
+        self.medium_combo = QComboBox()
+        self.medium_combo.addItems(
+            ["photograph", "illustration", "3d_render", "painting", "graphic_design", "mixed-media digital collage"]
+        )
+        self.palette_editor = PaletteEditor(limit=16)
+        self.install_translate_menu(self.aesthetics_edit)
+        self.install_translate_menu(self.lighting_edit)
+        self.install_translate_menu(self.photo_edit)
+        self.install_translate_menu(self.art_style_edit)
+        self.install_translate_menu(self.palette_editor.line_edit)
+
+        form.addRow(tr("style.aesthetics"), self.aesthetics_edit)
+        form.addRow(tr("style.lighting"), self.lighting_edit)
+        self.photo_row_label = QLabel(tr("style.photo_field"))
+        self.art_row_label = QLabel(tr("style.art_style"))
+        form.addRow(self.photo_row_label, self.photo_edit)
+        form.addRow(self.art_row_label, self.art_style_edit)
+        form.addRow(tr("style.medium"), self.medium_combo)
+        form.addRow(tr("style.palette"), self.palette_editor)
+        layout.addLayout(form)
+
+        for widget in [self.aesthetics_edit, self.lighting_edit, self.photo_edit, self.art_style_edit]:
+            widget.textChanged.connect(self.update_output)
+        self.medium_combo.currentTextChanged.connect(self.update_output)
+        self.palette_editor.changed.connect(self.update_output)
+        self.editor_layout.addWidget(box)
+
+    def _build_composition(self):
+        box = QGroupBox(tr("grp.composition"))
+        layout = QVBoxLayout(box)
+        self.background_text = QTextEdit()
+        self.background_text.setMinimumHeight(95)
+        self.background_text.setPlaceholderText(tr("comp.background_placeholder"))
+        self.background_text.textChanged.connect(self.update_output)
+        self.install_translate_menu(self.background_text)
+        layout.addWidget(QLabel(tr("comp.background")))
+        layout.addWidget(self.background_text)
+
+        body = QHBoxLayout()
+        body.setSpacing(14)
+        self.element_list = QListWidget()
+        self.element_list.currentRowChanged.connect(self.select_element)
+        self.element_list.setMinimumWidth(280)
+        left = QVBoxLayout()
+        add_row = QHBoxLayout()
+        add_button = QPushButton(tr("comp.add_element"))
+        add_button.setObjectName("PrimaryButton")
+        add_button.clicked.connect(lambda: self.add_element())
+        template_button = QPushButton(tr("tb.template"))
+        template_button.clicked.connect(self.add_from_template)
+        add_row.addWidget(add_button, 1)
+        add_row.addWidget(template_button)
+        left.addLayout(add_row)
+        left.addWidget(self.element_list, 1)
+        ops_row = QHBoxLayout()
+        for label, callback in [
+            (tr("tb.duplicate"), self.duplicate_element),
+            (tr("tb.move_up"), lambda: self.move_element(-1)),
+            (tr("tb.move_down"), lambda: self.move_element(1)),
+        ]:
+            button = QPushButton(label)
+            button.clicked.connect(callback)
+            ops_row.addWidget(button)
+        left.addLayout(ops_row)
+        remove_button = QPushButton(tr("comp.remove_element"))
+        remove_button.clicked.connect(self.delete_element)
+        left.addWidget(remove_button)
+        body.addLayout(left, 1)
+
+        right = QVBoxLayout()
+        self._build_element_form(right)
+
+        ref_row = QHBoxLayout()
+        for label, callback in [
+            (tr("canvas.load_ref"), self.load_reference_image),
+            (tr("canvas.paste_ref"), self.paste_reference_image),
+            (tr("canvas.clear_ref"), self.clear_reference_image),
+        ]:
+            button = QPushButton(label)
+            button.clicked.connect(callback)
+            ref_row.addWidget(button)
+        right.addLayout(ref_row)
+
+        zoom_row = QHBoxLayout()
+        zoom_row.addWidget(QLabel(tr("canvas.zoom")))
+        self.zoom_slider = QSlider(Qt.Orientation.Horizontal)
+        self.zoom_slider.setRange(50, 300)
+        self.zoom_slider.setValue(100)
+        self.zoom_label = QLabel("100%")
+        self.zoom_slider.valueChanged.connect(self._on_zoom_changed)
+        zoom_row.addWidget(self.zoom_slider, 1)
+        zoom_row.addWidget(self.zoom_label)
+        right.addLayout(zoom_row)
+
+        self.canvas = BBoxCanvas()
+        self.canvas.set_theme(self.theme)
+        self.canvas.selected.connect(self.select_element)
+        self.canvas.bbox_changed.connect(self.update_bbox_from_canvas)
+        canvas_scroll = QScrollArea()
+        canvas_scroll.setWidgetResizable(True)
+        canvas_scroll.setWidget(self.canvas)
+        canvas_scroll.setMinimumHeight(380)
+        right.addWidget(canvas_scroll)
+        hint = QLabel(tr("comp.hint"))
+        hint.setWordWrap(True)
+        hint.setStyleSheet(f"color:{THEMES[self.theme]['muted']};background:transparent;")
+        right.addWidget(hint)
+        body.addLayout(right, 2)
+        layout.addLayout(body)
+        self.editor_layout.addWidget(box)
+
+    def _build_element_form(self, parent_layout):
+        form_box = QFrame()
+        form_layout = QFormLayout(form_box)
+        self.element_type = QComboBox()
+        self.element_type.addItems(["obj", "text"])
+        self.element_label = QLineEdit()
+        self.element_text = QLineEdit()
+        self.element_desc = QTextEdit()
+        self.element_desc.setMinimumHeight(90)
+        self.element_palette = PaletteEditor(limit=5)
+        self.install_translate_menu(self.element_label)
+        self.install_translate_menu(self.element_text)
+        self.install_translate_menu(self.element_desc)
+        self.install_translate_menu(self.element_palette.line_edit)
+        self.use_bbox = QCheckBox(tr("el.use_bbox"))
+        self.use_bbox.setChecked(True)
+        self.bbox_spins = []
+        bbox_layout = QHBoxLayout()
+        for name in ["Y min", "X min", "Y max", "X max"]:
+            spin = QSpinBox()
+            spin.setRange(0, 1000)
+            spin.setValue(200 if "min" in name else 800)
+            spin.setPrefix(f"{name}: ")
+            self.bbox_spins.append(spin)
+            bbox_layout.addWidget(spin)
+
+        form_layout.addRow(tr("el.type"), self.element_type)
+        form_layout.addRow(tr("el.label"), self.element_label)
+        form_layout.addRow(tr("el.text"), self.element_text)
+        form_layout.addRow(tr("el.description"), self.element_desc)
+        form_layout.addRow(tr("el.palette"), self.element_palette)
+        form_layout.addRow("", self.use_bbox)
+        form_layout.addRow(tr("el.bbox"), bbox_layout)
+        parent_layout.addWidget(form_box)
+
+        for widget in [self.element_type, self.element_label, self.element_text, self.use_bbox, *self.bbox_spins]:
+            signal = (
+                widget.currentTextChanged
+                if isinstance(widget, QComboBox)
+                else widget.textChanged
+                if isinstance(widget, QLineEdit)
+                else widget.stateChanged
+                if isinstance(widget, QCheckBox)
+                else widget.valueChanged
+            )
+            signal.connect(self.save_element_form)
+        self.element_desc.textChanged.connect(self.save_element_form)
+        self.element_palette.changed.connect(self.save_element_form)
+
+    def _build_output(self, layout):
+        self.output_tabs = QTabWidget()
+        layout.addWidget(self.output_tabs, 1)
+
+        # --- JSON tab ---
+        json_tab = QWidget()
+        json_layout = QVBoxLayout(json_tab)
+        json_layout.setContentsMargins(0, 8, 0, 0)
+        top = QHBoxLayout()
+        title = QLabel(tr("out.title"))
+        title.setStyleSheet("font-size:16px;font-weight:700;background:transparent;")
+        self.pretty_radio = QRadioButton(tr("out.pretty"))
+        self.compact_radio = QRadioButton(tr("out.compact"))
+        self.pretty_radio.setChecked(True)
+        self.pretty_radio.toggled.connect(self.update_output)
+        top.addWidget(title)
+        top.addStretch()
+        top.addWidget(self.pretty_radio)
+        top.addWidget(self.compact_radio)
+        json_layout.addLayout(top)
+
+        self.output_text = QPlainTextEdit()
+        self.output_text.setReadOnly(True)
+        self.output_text.setLineWrapMode(QPlainTextEdit.LineWrapMode.NoWrap)
+        self.output_text.setSizePolicy(QSizePolicy.Policy.Expanding, QSizePolicy.Policy.Expanding)
+        json_layout.addWidget(self.output_text, 1)
+
+        actions = QHBoxLayout()
+        copy_compact = QPushButton(tr("out.copy_compact"))
+        copy_compact.clicked.connect(self.copy_compact_json)
+        save = QPushButton(tr("out.save_json_btn"))
+        save.clicked.connect(self.save_json)
+        actions.addWidget(copy_compact)
+        actions.addWidget(save)
+        actions.addStretch()
+        json_layout.addLayout(actions)
+
+        self.validation_list = QListWidget()
+        self.validation_list.setMaximumHeight(160)
+        self.validation_list.itemClicked.connect(self._on_validation_clicked)
+        json_layout.addWidget(self.validation_list)
+        self.output_tabs.addTab(json_tab, tr("tab.json"))
+
+        # --- Result tab (generated image from ComfyUI, item 14) ---
+        result_tab = QWidget()
+        result_layout = QVBoxLayout(result_tab)
+        result_layout.setContentsMargins(0, 8, 0, 0)
+        self.result_label = QLabel(tr("result.empty"))
+        self.result_label.setAlignment(Qt.AlignmentFlag.AlignCenter)
+        self.result_label.setStyleSheet(
+            "background:palette(base);border:1px solid palette(mid);border-radius:8px;"
+        )
+        self.result_label.setSizePolicy(QSizePolicy.Policy.Expanding, QSizePolicy.Policy.Expanding)
+        result_layout.addWidget(self.result_label, 1)
+        result_actions = QHBoxLayout()
+        self.result_save_lib = QPushButton(tr("result.save_lib"))
+        self.result_save_lib.clicked.connect(lambda: self._save_generated_to_library(self._last_generated))
+        self.result_save_file = QPushButton(tr("result.save_file"))
+        self.result_save_file.clicked.connect(self._save_generated_to_file)
+        self.result_save_lib.setEnabled(False)
+        self.result_save_file.setEnabled(False)
+        result_actions.addWidget(self.result_save_lib)
+        result_actions.addWidget(self.result_save_file)
+        result_actions.addStretch()
+        result_layout.addLayout(result_actions)
+        self.output_tabs.addTab(result_tab, tr("tab.result"))
+        self._last_generated = None
+
+    def _style_mode_changed(self):
+        photo_mode = self.photo_radio.isChecked()
+        self.photo_edit.setVisible(photo_mode)
+        self.photo_row_label.setVisible(photo_mode)
+        self.art_style_edit.setVisible(not photo_mode)
+        self.art_row_label.setVisible(not photo_mode)
+        if self._loading:
+            return
+        if photo_mode:
+            self.medium_combo.setCurrentText("photograph")
+        elif self.medium_combo.currentText() == "photograph":
+            self.medium_combo.setCurrentText("illustration")
+        self.update_output()
+
+    def style_mode(self):
+        return "photo" if self.photo_radio.isChecked() else "art"
+
+    def current_caption(self):
+        caption = {}
+        high = self.high_text.toPlainText().strip()
+        if high:
+            caption["high_level_description"] = high
+
+        style = {}
+        if self.aesthetics_edit.text().strip():
+            style["aesthetics"] = self.aesthetics_edit.text().strip()
+        if self.lighting_edit.text().strip():
+            style["lighting"] = self.lighting_edit.text().strip()
+        if self.style_mode() == "photo":
+            if self.photo_edit.text().strip():
+                style["photo"] = self.photo_edit.text().strip()
+            if self.medium_combo.currentText().strip():
+                style["medium"] = self.medium_combo.currentText().strip()
+        else:
+            if self.medium_combo.currentText().strip():
+                style["medium"] = self.medium_combo.currentText().strip()
+            if self.art_style_edit.text().strip():
+                style["art_style"] = self.art_style_edit.text().strip()
+        if self.palette_editor.colors():
+            style["color_palette"] = self.palette_editor.colors()
+        if style:
+            caption["style_description"] = style
+
+        caption["compositional_deconstruction"] = {
+            "background": self.background_text.toPlainText().strip(),
+            "elements": [self.ordered_element(element) for element in self.elements],
+        }
+        return caption
+
+    def ordered_element(self, element):
+        item = {"type": element["type"]}
+        if element.get("use_bbox"):
+            item["bbox"] = [int(value) for value in element["bbox"]]
+        if element["type"] == "text":
+            item["text"] = element.get("text", "").strip()
+        item["desc"] = element.get("desc", "").strip()
+        colors = parse_palette(element.get("palette", ""), 5)
+        if colors:
+            item["color_palette"] = colors
+        return item
+
+    def update_output(self):
+        if self._loading:
+            return
+        caption = self.current_caption()
+        if self.compact_radio.isChecked():
+            text = json.dumps(caption, ensure_ascii=False, separators=(",", ":"))
+        else:
+            text = json.dumps(caption, ensure_ascii=False, indent=2)
+        self.output_text.setPlainText(text)
+        self._populate_validation(self.validate_caption(caption))
+        self.canvas.set_data(self.elements, self.selected_index)
+        self._push_history()
+        self._save_draft(caption)
+
+    def _populate_validation(self, messages):
+        colors = {"ok": "#2E8B57", "warn": "#B8860B", "bad": THEMES[self.theme]["error"]}
+        self.validation_list.clear()
+        for kind, message, element_index in messages:
+            item = QListWidgetItem(f"[{kind.upper()}] {message}")
+            item.setForeground(QColor(colors.get(kind, THEMES[self.theme]["text"])))
+            item.setData(Qt.ItemDataRole.UserRole, element_index)
+            self.validation_list.addItem(item)
+
+    def _on_validation_clicked(self, item):
+        index = item.data(Qt.ItemDataRole.UserRole)
+        if index is not None and 0 <= index < len(self.elements):
+            self.select_element(index)
+
+    def validate_caption(self, caption):
+        """Return a list of (kind, message, element_index_or_None) tuples."""
+        messages = []
+        style = caption.get("style_description", {})
+        comp = caption["compositional_deconstruction"]
+        if not caption.get("high_level_description"):
+            messages.append(("warn", tr("val.no_high"), None))
+        if not comp.get("background"):
+            messages.append(("bad", tr("val.bg_required"), None))
+        if not comp.get("elements"):
+            messages.append(("bad", tr("val.add_element"), None))
+        if style:
+            missing = [key for key in ["aesthetics", "lighting", "medium"] if not style.get(key)]
+            if missing:
+                messages.append(("bad", tr("val.style_missing").format(fields=", ".join(missing)), None))
+            if bool(style.get("photo")) == bool(style.get("art_style")):
+                messages.append(("bad", tr("val.photo_or_art"), None))
+            for color in style.get("color_palette", []):
+                if not HEX_RE.match(color):
+                    messages.append(("bad", tr("val.hex_upper").format(color=color), None))
+        for index, element in enumerate(comp.get("elements", []), start=1):
+            ei = index - 1
+            title = element.get("text") or tr("val.element_word").format(index=index)
+            if element["type"] == "text" and not element.get("text"):
+                messages.append(("bad", tr("val.text_literal").format(title=title), ei))
+            if not element.get("desc"):
+                messages.append(("bad", tr("val.desc_required").format(title=title), ei))
+            if "bbox" in element:
+                y1, x1, y2, x2 = element["bbox"]
+                if y2 <= y1 or x2 <= x1:
+                    messages.append(("bad", tr("val.bbox_order").format(title=title), ei))
+            for color in element.get("color_palette", []):
+                if not HEX_RE.match(color):
+                    messages.append(("bad", tr("val.el_hex").format(title=title, color=color), ei))
+        if not any(kind == "bad" for kind, _message, _idx in messages):
+            messages.insert(0, ("ok", tr("val.ok"), None))
+        return messages
+
+    def add_element(self, element=None):
+        element = element or {}
+        normalized = {
+            "type": element.get("type", "obj"),
+            "label": element.get("label") or element.get("text") or f"{tr('el.element')} {len(self.elements) + 1}",
+            "text": element.get("text", ""),
+            "desc": element.get("desc", ""),
+            "palette": palette_text(element.get("color_palette", []))
+            if isinstance(element.get("color_palette"), list)
+            else element.get("palette", ""),
+            "use_bbox": "bbox" in element or element.get("use_bbox", True),
+            "bbox": element.get("bbox", [200, 200, 800, 800]),
+        }
+        self.elements.append(normalized)
+        self.refresh_elements(len(self.elements) - 1)
+
+    def delete_element(self):
+        if self.selected_index is None:
+            return
+        del self.elements[self.selected_index]
+        next_index = min(self.selected_index, len(self.elements) - 1) if self.elements else None
+        self.refresh_elements(next_index)
+
+    def refresh_elements(self, selected_index=None):
+        self.element_list.blockSignals(True)
+        self.element_list.clear()
+        for index, element in enumerate(self.elements, start=1):
+            title = element.get("text") or element.get("label") or element.get("desc", "")[:32] or f"{tr('el.element')} {index}"
+            self.element_list.addItem(QListWidgetItem(f"{index}. {element['type']} - {title}"))
+        self.element_list.blockSignals(False)
+        self.selected_index = selected_index
+        if selected_index is not None and selected_index >= 0:
+            self.element_list.setCurrentRow(selected_index)
+        self.load_element_form()
+        self.update_output()
+
+    def select_element(self, row):
+        if row < 0:
+            self.selected_index = None
+        else:
+            self.selected_index = row
+            if self.element_list.currentRow() != row:
+                self.element_list.setCurrentRow(row)
+        self.load_element_form()
+        self.update_output()
+
+    def load_element_form(self):
+        self._loading = True
+        enabled = self.selected_index is not None and bool(self.elements)
+        for widget in [
+            self.element_type,
+            self.element_label,
+            self.element_text,
+            self.element_desc,
+            self.element_palette,
+            self.use_bbox,
+            *self.bbox_spins,
+        ]:
+            widget.setEnabled(enabled)
+        if enabled:
+            element = self.elements[self.selected_index]
+            self.element_type.setCurrentText(element["type"])
+            self.element_label.setText(element.get("label", ""))
+            self.element_text.setText(element.get("text", ""))
+            self.element_desc.setPlainText(element.get("desc", ""))
+            self.element_palette.set_text(element.get("palette", ""))
+            self.use_bbox.setChecked(element.get("use_bbox", True))
+            for spin, value in zip(self.bbox_spins, element.get("bbox", [200, 200, 800, 800])):
+                spin.setValue(int(value))
+        self._loading = False
+
+    def save_element_form(self):
+        if self._loading or self.selected_index is None:
+            return
+        self.elements[self.selected_index] = {
+            "type": self.element_type.currentText(),
+            "label": self.element_label.text().strip(),
+            "text": self.element_text.text().strip(),
+            "desc": self.element_desc.toPlainText().strip(),
+            "palette": self.element_palette.text(),
+            "use_bbox": self.use_bbox.isChecked(),
+            "bbox": [spin.value() for spin in self.bbox_spins],
+        }
+        current = self.selected_index
+        self.element_list.blockSignals(True)
+        item = self.element_list.item(current)
+        if item:
+            element = self.elements[current]
+            title = element.get("text") or element.get("label") or element.get("desc", "")[:32] or f"{tr('el.element')} {current + 1}"
+            item.setText(f"{current + 1}. {element['type']} - {title}")
+        self.element_list.blockSignals(False)
+        self.update_output()
+
+    def update_bbox_from_canvas(self, index, bbox):
+        if index < 0 or index >= len(self.elements):
+            return
+        self.elements[index]["use_bbox"] = True
+        self.elements[index]["bbox"] = bbox
+        if self.selected_index != index:
+            self.select_element(index)
+        self._loading = True
+        for spin, value in zip(self.bbox_spins, bbox):
+            spin.setValue(int(value))
+        self.use_bbox.setChecked(True)
+        self._loading = False
+        self.update_output()
+
+    # --- Element operations (items 3, 4, 12) ----------------------------
+    def duplicate_element(self):
+        if self.selected_index is None:
+            return
+        clone = copy.deepcopy(self.elements[self.selected_index])
+        clone["label"] = f"{clone.get('label', '')} copy".strip()
+        self.elements.insert(self.selected_index + 1, clone)
+        self.refresh_elements(self.selected_index + 1)
+
+    def move_element(self, delta):
+        if self.selected_index is None:
+            return
+        new_index = self.selected_index + delta
+        if new_index < 0 or new_index >= len(self.elements):
+            return
+        items = self.elements
+        items[self.selected_index], items[new_index] = items[new_index], items[self.selected_index]
+        self.refresh_elements(new_index)
+
+    def add_from_template(self):
+        names = list(ELEMENT_TEMPLATES.keys())
+        name, ok = QInputDialog.getItem(
+            self, tr("tpl.choose_title"), tr("tpl.choose_label"), names, 0, False
+        )
+        if not ok or not name:
+            return
+        template = copy.deepcopy(ELEMENT_TEMPLATES[name])
+        template.setdefault("use_bbox", True)
+        self.add_element(template)
+
+    # --- Reference image + zoom (item 5) --------------------------------
+    def load_reference_image(self):
+        path, _filter = QFileDialog.getOpenFileName(
+            self, tr("canvas.load_ref"), "", tr("prev.filter")
+        )
+        if not path:
+            return
+        pixmap = QPixmap(path)
+        if pixmap.isNull():
+            QMessageBox.warning(self, tr("canvas.load_ref"), tr("canvas.ref_load_fail"))
+            return
+        self.canvas.set_reference(pixmap)
+
+    def paste_reference_image(self):
+        image = QGuiApplication.clipboard().image()
+        if image.isNull():
+            QMessageBox.information(self, tr("canvas.paste_ref"), tr("libd.no_clipboard_image"))
+            return
+        self.canvas.set_reference(QPixmap.fromImage(image))
+
+    def clear_reference_image(self):
+        self.canvas.set_reference(None)
+
+    def _on_zoom_changed(self, value):
+        self.zoom_label.setText(f"{value}%")
+        self.canvas.set_zoom(value)
+
+    # --- Draft autosave (item 2) ----------------------------------------
+    def _save_draft(self, caption=None):
+        if self._loading:
+            return
+        try:
+            with open(DRAFT_FILE, "w", encoding="utf-8") as handle:
+                json.dump(caption if caption is not None else self.current_caption(),
+                          handle, ensure_ascii=False, indent=2)
+        except OSError:
+            pass
+
+    def _restore_draft(self):
+        if not DRAFT_FILE.exists():
+            return False
+        try:
+            with open(DRAFT_FILE, "r", encoding="utf-8") as handle:
+                caption = json.load(handle)
+        except (OSError, json.JSONDecodeError):
+            return False
+        if not isinstance(caption, dict) or not caption.get("compositional_deconstruction"):
+            return False
+        if QMessageBox.question(
+            self, tr("draft.restore_title"), tr("draft.restore_q")
+        ) == QMessageBox.StandardButton.Yes:
+            self.load_caption(caption)
+            return True
+        return False
+
+    def closeEvent(self, event):
+        self._save_draft()
+        if self._gen_thread is not None and self._gen_thread.isRunning():
+            self._gen_thread.cancel()
+            self._gen_thread.wait(2000)
+        super().closeEvent(event)
+
+    # --- ComfyUI (item 14) ----------------------------------------------
+    def open_comfy_settings(self):
+        dialog = ComfySettingsDialog(self.settings, self)
+        if dialog.exec() == QDialog.DialogCode.Accepted:
+            self.settings.update(dialog.values())
+            save_settings(self.settings)
+            QMessageBox.information(self, tr("set.title"), tr("set.saved"))
+
+    def _missing_deps_report(self, missing):
+        sections = [
+            ("nodes", "comfy.missing_nodes"), ("unet", "comfy.missing_unet"),
+            ("vae", "comfy.missing_vae"), ("clip", "comfy.missing_clip"),
+            ("clip_gguf", "comfy.missing_clip_gguf"), ("samplers", "comfy.missing_samplers"),
+        ]
+        lines = []
+        for key, tkey in sections:
+            if missing.get(key):
+                lines.append(tr(tkey).format(items=", ".join(missing[key])))
+        return lines
+
+    def check_comfy(self):
+        progress = QProgressDialog(tr("comfy.checking"), tr("common.cancel"), 0, 0, self)
+        progress.setWindowTitle(tr("comfy.check_title"))
+        progress.setMinimumDuration(0)
+        progress.setValue(0)
+        QApplication.processEvents()
+        try:
+            missing = check_comfy_dependencies(self.settings)
+        except ComfyError as error:
+            progress.close()
+            QMessageBox.warning(
+                self, tr("comfy.check_title"),
+                tr("comfy.unreachable").format(url=comfy_base_url(self.settings), err=error),
+            )
+            return None
+        progress.close()
+        lines = self._missing_deps_report(missing)
+        if not lines:
+            QMessageBox.information(self, tr("comfy.check_title"), tr("comfy.all_ok"))
+        else:
+            QMessageBox.warning(
+                self, tr("comfy.check_title"),
+                tr("comfy.missing_header") + "\n\n" + "\n".join(lines),
+            )
+        return missing
+
+    def generate_in_comfy(self):
+        if not WORKFLOW_FILE.exists():
+            QMessageBox.critical(self, tr("comfy.gen_title"),
+                                 tr("comfy.workflow_missing").format(path=WORKFLOW_FILE))
+            return
+        missing = self.check_comfy()
+        if missing is None:
+            return  # server unreachable, already reported
+        if any(missing.values()):
+            if QMessageBox.question(
+                self, tr("comfy.gen_title"), tr("comfy.deps_missing_continue")
+            ) != QMessageBox.StandardButton.Yes:
+                return
+        try:
+            with open(WORKFLOW_FILE, "r", encoding="utf-8") as handle:
+                workflow = json.load(handle)
+        except (OSError, json.JSONDecodeError) as error:
+            QMessageBox.critical(self, tr("comfy.gen_title"), tr("comfy.gen_fail").format(err=error))
+            return
+
+        caption = json.dumps(self.current_caption(), ensure_ascii=False, separators=(",", ":"))
+        seed = uuid.uuid4().int % (2 ** 31)
+        self._gen_progress = QProgressDialog(tr("comfy.generating"), tr("common.cancel"), 0, 0, self)
+        self._gen_progress.setWindowTitle(tr("comfy.gen_title"))
+        self._gen_progress.setMinimumDuration(0)
+        self._gen_progress.setValue(0)
+
+        self._gen_thread = GenerationThread(self.settings, workflow, caption, seed, self)
+        self._gen_thread.finished_ok.connect(self._on_generation_done)
+        self._gen_thread.failed.connect(self._on_generation_failed)
+        self._gen_progress.canceled.connect(self._gen_thread.cancel)
+        self._gen_thread.start()
+
+    def _on_generation_failed(self, message):
+        if getattr(self, "_gen_progress", None):
+            self._gen_progress.close()
+        if message != "cancelled":
+            QMessageBox.warning(self, tr("comfy.gen_title"), tr("comfy.gen_fail").format(err=message))
+
+    def _on_generation_done(self, data):
+        if getattr(self, "_gen_progress", None):
+            self._gen_progress.close()
+        self._last_generated = data
+        pixmap = QPixmap()
+        pixmap.loadFromData(data)
+        if not pixmap.isNull():
+            self._result_pixmap = pixmap
+            self._render_result()
+            self.canvas.set_reference(pixmap)
+        self.result_save_lib.setEnabled(True)
+        self.result_save_file.setEnabled(True)
+        # Bring the generated image to the foreground (item: get image into the app).
+        self.output_tabs.setCurrentIndex(1)
+
+    def _render_result(self):
+        pixmap = getattr(self, "_result_pixmap", None)
+        if pixmap is None or pixmap.isNull():
+            return
+        target = self.result_label.size()
+        self.result_label.setPixmap(
+            pixmap.scaled(target, Qt.AspectRatioMode.KeepAspectRatio, Qt.TransformationMode.SmoothTransformation)
+        )
+
+    def resizeEvent(self, event):
+        super().resizeEvent(event)
+        self._render_result()
+
+    def _save_generated_to_file(self):
+        if not self._last_generated:
+            return
+        path, _filter = QFileDialog.getSaveFileName(
+            self, tr("result.save_file"), "ideogram-result.png", tr("result.png_filter")
+        )
+        if not path:
+            return
+        try:
+            with open(path, "wb") as handle:
+                handle.write(self._last_generated)
+        except OSError as error:
+            QMessageBox.warning(self, tr("comfy.gen_title"), tr("comfy.gen_fail").format(err=error))
+            return
+        QMessageBox.information(self, tr("comfy.gen_title"), tr("result.saved_file").format(path=path))
+
+    def _save_generated_to_library(self, image_data):
+        caption = self.current_caption()
+        default = caption.get("high_level_description", "")[:48].strip() or tr("lib.untitled")
+        name, ok = QInputDialog.getText(self, tr("tb.save_library"), tr("lib.name_prompt"), text=default)
+        if not ok or not name.strip():
+            return
+        entries = load_library()
+        now = datetime.now().isoformat(timespec="seconds")
+        entry = {
+            "id": uuid.uuid4().hex, "name": name.strip(), "created": now, "updated": now,
+            "preview": None, "tags": [], "caption": caption,
+        }
+        try:
+            PREVIEW_DIR.mkdir(parents=True, exist_ok=True)
+            target_name = f"{entry['id']}.png"
+            with open(PREVIEW_DIR / target_name, "wb") as handle:
+                handle.write(image_data)
+            entry["preview"] = target_name
+        except OSError:
+            pass
+        entries.append(entry)
+        try:
+            save_library(entries)
+        except OSError as error:
+            QMessageBox.critical(self, tr("tb.library"), tr("lib.save_fail").format(err=error))
+            return
+        self._library_entry_id = entry["id"]
+        QMessageBox.information(self, tr("tb.library"), tr("lib.saved").format(name=entry["name"]))
+
+    def load_caption(self, caption, mark_history=False):
+        self._loading = True
+        self.high_text.setPlainText(caption.get("high_level_description", ""))
+        style = caption.get("style_description", {})
+        self.photo_radio.setChecked("art_style" not in style)
+        self.art_radio.setChecked("art_style" in style)
+        self.aesthetics_edit.setText(style.get("aesthetics", ""))
+        self.lighting_edit.setText(style.get("lighting", ""))
+        self.photo_edit.setText(style.get("photo", ""))
+        self.art_style_edit.setText(style.get("art_style", ""))
+        self.medium_combo.setCurrentText(style.get("medium", "photograph"))
+        self.palette_editor.set_colors(style.get("color_palette", []))
+        comp = caption.get("compositional_deconstruction", {})
+        self.background_text.setPlainText(comp.get("background", ""))
+        self.elements = []
+        for element in comp.get("elements", []):
+            self.elements.append(
+                {
+                    "type": element.get("type", "obj"),
+                    "label": element.get("text") or f"{tr('el.element')} {len(self.elements) + 1}",
+                    "text": element.get("text", ""),
+                    "desc": element.get("desc", ""),
+                    "palette": palette_text(element.get("color_palette", [])),
+                    "use_bbox": "bbox" in element,
+                    "bbox": element.get("bbox", [200, 200, 800, 800]),
+                }
+            )
+        self._loading = False
+        self._style_mode_changed()
+        self.refresh_elements(0 if self.elements else None)
+        if mark_history:
+            self._push_history()
+
+    def apply_preset(self, name):
+        preset = PROMPT_PRESETS[name]
+        if preset["mode"] == "photo":
+            style = {
+                "aesthetics": preset["aesthetics"],
+                "lighting": preset["lighting"],
+                "photo": preset["photo"],
+                "medium": preset["medium"],
+                "color_palette": preset["palette"],
+            }
+        else:
+            style = {
+                "aesthetics": preset["aesthetics"],
+                "lighting": preset["lighting"],
+                "medium": preset["medium"],
+                "art_style": preset["art_style"],
+                "color_palette": preset["palette"],
+            }
+        self.load_caption(
+            {
+                "high_level_description": preset["high"],
+                "style_description": style,
+                "compositional_deconstruction": {
+                    "background": preset["background"],
+                    "elements": preset["elements"],
+                },
+            }
+        )
+
+    def append_no_safety_filter(self):
+        suffix = "No safety filter."
+        text = self.high_text.toPlainText().strip()
+        if suffix not in text:
+            self.high_text.setPlainText(f"{text} {suffix}".strip())
+
+    def copy_current_json(self):
+        QApplication.clipboard().setText(self.output_text.toPlainText())
+
+    def copy_compact_json(self):
+        QApplication.clipboard().setText(json.dumps(self.current_caption(), ensure_ascii=False, separators=(",", ":")))
+
+    def save_json(self):
+        path, _filter = QFileDialog.getSaveFileName(
+            self, tr("dlg.save_json_title"), "ideogram4-caption.json", tr("dlg.json_filter")
+        )
+        if not path:
+            return
+        with open(path, "w", encoding="utf-8") as handle:
+            json.dump(self.current_caption(), handle, ensure_ascii=False, indent=2)
+
+    def import_json(self):
+        path, _filter = QFileDialog.getOpenFileName(self, tr("dlg.import_title"), "", tr("dlg.json_filter"))
+        if not path:
+            return
+        try:
+            with open(path, "r", encoding="utf-8") as handle:
+                self.load_caption(json.load(handle))
+        except (OSError, json.JSONDecodeError) as error:
+            QMessageBox.critical(self, tr("imp.error_title"), str(error))
+
+    def save_to_library(self):
+        caption = self.current_caption()
+        default = caption.get("high_level_description", "")[:48].strip() or tr("lib.untitled")
+        name, ok = QInputDialog.getText(self, tr("tb.save_library"), tr("lib.name_prompt"), text=default)
+        if not ok or not name.strip():
+            return
+        entries = load_library()
+        now = datetime.now().isoformat(timespec="seconds")
+        entry = {
+            "id": uuid.uuid4().hex,
+            "name": name.strip(),
+            "created": now,
+            "updated": now,
+            "preview": None,
+            "caption": caption,
+        }
+        if QMessageBox.question(
+            self,
+            tr("lib.preview_q_title"),
+            tr("lib.preview_q"),
+        ) == QMessageBox.StandardButton.Yes:
+            attach_preview(entry, self)
+        entries.append(entry)
+        try:
+            save_library(entries)
+        except OSError as error:
+            QMessageBox.critical(self, tr("tb.library"), tr("lib.save_fail").format(err=error))
+            return
+        self._library_entry_id = entry["id"]
+        QMessageBox.information(self, tr("tb.library"), tr("lib.saved").format(name=entry["name"]))
+
+    def overwrite_in_library(self):
+        """Update the library entry the current prompt was loaded from (item 7)."""
+        entries = load_library()
+        entry = next((e for e in entries if e.get("id") == self._library_entry_id), None)
+        if entry is None:
+            # Nothing to overwrite — fall back to saving a new entry.
+            self.save_to_library()
+            return
+        entry["caption"] = self.current_caption()
+        entry["updated"] = datetime.now().isoformat(timespec="seconds")
+        try:
+            save_library(entries)
+        except OSError as error:
+            QMessageBox.critical(self, tr("tb.library"), tr("lib.save_fail").format(err=error))
+            return
+        QMessageBox.information(self, tr("tb.library"), tr("lib.saved").format(name=entry.get("name", "")))
+
+    def open_library(self):
+        entries = load_library()
+        dialog = LibraryDialog(entries, self)
+        if dialog.exec() == QDialog.DialogCode.Accepted and dialog.selected_caption is not None:
+            self.load_caption(dialog.selected_caption)
+            self._library_entry_id = dialog.selected_id
+
+
+def main():
+    app = QApplication(sys.argv)
+    window = PromptBuilder()
+    window.show()
+    sys.exit(app.exec())
+
+
+if __name__ == "__main__":
+    main()
diff --git a/ru-white.png b/ru-white.png
new file mode 100644
index 0000000..ac3b55f
Binary files /dev/null and b/ru-white.png differ