ComfyUI Node
toobusy Ideogram Prompt Polish
A ComfyUI node in toobusy/Plan with 13 inputs and 2 outputs.
toobusy Ideogram Prompt Polish
- clip
- image
- ideogram_json
- llm_raw
◄scene노을 지는 해변에서 빨간 원피스를 입은 여자가 카메라를 등지고 서 있다.►
◄style_modeLiteral►
◄languageAuto►
◄preserve_intenttrue►
◄fill_missing_fieldstrue►
◄seed1►
◄existing_layout_json►
◄image_instruction_modeAnalyze image literally►
◄analysis_modebalanced►
◄debug_rawfalse►
◄release_clip_after_runtrue►
Categorytoobusy/Plan
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| clip | CLIP | — | |
| scene | STRING | 노을 지는 해변에서 빨간 원피스를 입은 여자가 카메라를 등지고 서 있다. | — |
| style_mode | COMBO | Literal | 5 options: Literal, Cinematic, Product, Character, Poster |
| language | COMBO | Auto | 3 options: Auto, Korean, English |
| preserve_intent | BOOLEAN | true | — |
| fill_missing_fields | BOOLEAN | true | — |
| seed | INT | 10–18446744073709550000 | — |
| existing_layout_jsonopt | STRING | — | |
| imageopt | IMAGE | Optional. Connect an image and a vision-capable text model (e.g. Gemma 4) to analyze the image INTO an Ideogram layout draft (reads text incl. Korean, finds elements + bboxes). Without an image this stays a scene-text polisher. Wire ideogram_json into the Layout Builder's imported_json, then press Pull. | |
| image_instruction_modeopt | COMBO | Analyze image literally | With an image connected: literal mode transcribes the source image; transform mode keeps the layout/bboxes but rewrites subject and text according to the Scene field. |
| analysis_modeopt | COMBO | balanced | Image analysis detail. fast: quick draft with few compact elements. balanced: default practical Layout Builder draft. detailed: slower, richer panel/UI/text/style analysis. |
| debug_rawopt | BOOLEAN | false | Off: return blank/truncated llm_raw to keep workflows lighter. On: return the full raw LLM response for debugging JSON failures. |
| release_clip_after_runopt | BOOLEAN | true | After creating the prompt, unload only this CLIP model from VRAM. Recommended before a large Ideogram model; unlike a global VRAM cleaner, other models and caches are preserved. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| ideogram_json | STRING | — |
| llm_raw | STRING | — |