IAMCCS IdeoTranslate

Ideogram 4's structured JSON prompt format is powerful but it's a chore to write by hand, and if you're building prompts with an LLM you probably have a big flat blob of text, not a nicely nested style_description object. IAMCCS IdeoTranslate sits exactly in that gap: it takes a structured prompt JSON - the kind your LLM prompt builder emits - and turns it into the design data this pack's Ideogram nodes actually consume.

Again, no API, no key, no network call. It's a JSON-to-JSON transform running locally. The author's own description calls it a "bridge node," which is the right mental model: one end speaks LLM-structured prompt, the other end speaks IAMCCS ideoboard.

How it works

Feed prompt_json_in anything shaped like the default - a high_level_description, a style_description with aesthetics/lighting/photo/medium/color_palette, and a compositional_deconstruction with background and elements. The node normalizes that into a design dict, then applies three optional passes, each toggled by a boolean:

• normalize_canvas_to_preset - snaps the canvas to a preset's proportions.
• auto_generate_text_boxes - creates text-element boxes so any copy in your prompt gets a real bounding box to live in.
• auto_palette_from_json - refreshes the scene color palette from the JSON's color_palette.

preset_override is the one knob you'll actually touch: auto lets the JSON decide, or you can force storyboard, poster, signage, screen_ui, or title_card. Handy when the same prompt text needs to be laid out as a poster in one branch and a storyboard panel in another - flip the combo, done.

Outputs

The four STRING outputs are all different views of the same design: design_data_json (the canonical IAMCCS design), ideoboard_package_json (the packaged form you hand to the Storyboard sheet node's ideotranslate_json input), prompt_json_out (back to Ideogram-style prompt JSON), and translation_summary (a readable summary of what changed). width/height report the normalized canvas.

Install & context

ComfyUI Manager → search IAMCCS, or clone the repo into custom_nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

Restart after. No extra dependencies, no model downloads.

The trap

If your LLM emits loosely-formed or free-text JSON, translation_summary is your friend - read it before wiring everything downstream. The node is defensive about missing fields but it can't invent a style_description that wasn't there; a minimal {"high_level_description": ...} will translate, but it'll translate to a pretty plain design. Start from the default JSON in the widget and you'll see the shape it wants.