IAMCCS StoryboardFrame V2 + Image Canvas i2i

IAMCCS_StoryboardFrameDesignerV2 is the upgraded storyboard node for Ideogram-style structured prompting: it turns a JSON design document - a full six-panel 2×3 cinematic contact sheet, with per-panel descriptions, colors, and layout - into the structured prompt that makes an Ideogram-class model render that sheet with text and layout intact. V2's new trick over the original is an Image Canvas i2i mode: instead of only producing a prompt, it can take an image canvas and run a low-denoise img2img pass over it so you can iterate on an existing board rather than regenerating from scratch.

Why structured JSON instead of just a prompt? Ideogram-class models are trained to respect a machine-readable layout - bbox coordinates, visible text, panel count - far better than a natural-language description. If you've ever watched an LLM-flavored prompt "put text in frame" and render gibberish, this is the discipline that fixes it: the model gets exact geometry, and the node keeps that geometry stable while you iterate on the content. (In this pack it feeds a larger Ideogram storyboard ecosystem; the IAMCCS_IdeoTranslate node produces these design packages, and this node consumes them.)

How it works

The design_data input is the editable working copy - a big JSON document holding the scene description, per-panel items (each with x, y, w, h, desc, and color_palette), the canvas size, and an i2i block. The optional ideoboard input can supply a newer design package; the ideoboard_input_signature guard exists so your manual canvas edits survive even while an IdeoTranslate input stays connected.

V2 adds the image side: i2i_enabled, i2i_denoise (default 0.28 - low, since you want to preserve the board and nudge details), and low_sigma_start_step (default 12, the step at which the low-sigma edit starts). Connect the optional source_image and V2 uses that IMAGE tensor directly as the i2i canvas guide instead of file-based layers. The output is a single IDEO_LINX - the pack's structured carrier that bundles the positive conditioning, the canvas image, i2i metadata, reference masks, and bounding boxes for the rest of the Ideogram generation chain (the scheduler hint in the design data even reminds you to use AutoCropGrid with 2 columns × 3 rows to crop the sheet back into panels).

Inputs and outputs

Required: clip, design_data (the JSON), ideoboard_input_signature, i2i_enabled, i2i_denoise, low_sigma_start_step. Optional: ideoboard, source_image. Output: IDEO_LINX.

Installing it

Part of the IAMCCS-nodes pack. ComfyUI Manager → search IAMCCS, or:

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

Restart ComfyUI; it's under IAMCCS/Cine/Ideogram. No pip requirements. The optional "Gemma assistant" in the design JSON is a local gemma4 text-encoder model path - purely optional, and it only activates if you enable it. README floor: ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8.

Where people get burned

• This node is one stage, not the whole pipeline. You still need the Ideogram-class model loaded, a CLIP to encode into, and the downstream grid-crop/decoding nodes. The IDEO_LINX output is designed for that chain - wiring it to a generic KSampler won't work as-is.
• i2i_denoise at 0.28 is a nudge, not a rewrite. Crank it toward 1.0 and the canvas guide stops being a guide - you get a new board that merely rhymes with the old one.
• Editing the JSON by hand is possible but easy to break; a malformed design_data silently falls back to defaults. Prefer the canvas UI if you're iterating, and treat direct JSON edits as an advanced move.