IAMCCS StoryboardFrame + TextInFrame Director

IAMCCS_StoryboardFrameDesigner is the original Ideogram storyboard node from the IAMCCS Cine pack: it takes a structured JSON design - a cinematic contact sheet with a fixed grid of panels, each with exact bounding boxes, descriptions, and color palettes - and produces the Ideogram-style prompt, the CLIP conditioning, and the canvas dimensions for generating it. The "TextInFrame Director" part of the display name is the selling point: Ideogram-class models are among the few that can render legible text inside an image, and this node leans on that by treating every panel's text and geometry as structured data instead of hoping the model infers it from prose.

If you're deciding between this and V2: V1 outputs classic positive CONDITIONING + prompt_json + width/height, and does not have V2's image-canvas i2i mode. V1 is the simpler, more composable one - it plays nice with a normal text-encoder/KSampler graph, whereas V2 bundles its state into the IDEO_LINX carrier for the bigger Ideogram chain. For a beginner, V1's outputs are easier to understand.

How it works

The design_data input is a big editable JSON document (schema iamccs.ideogram_storyboard_frame_designer). It describes the scene (high-level description, aesthetics, lighting, photo treatment, color palette), the background (e.g. "Clean storyboard contact sheet with exactly two columns and three rows"), and an items array where each panel carries x/y/w/h coordinates, a label, optional visible text, and a full panel desc. The node normalizes that design, builds the Ideogram prompt payload, encodes it into conditioning with your CLIP, and reports canvas width/height.

The optional ideoboard input accepts a design package from IAMCCS_IdeoTranslate; the ideoboard_input_signature is an internal guard so your manual edits to design_data survive when that input stays connected - a detail that prevents "I edited the board and the next run overwrote it" frustration. There's also a json_override mechanism in the design data, and the gemma_json_request output is the hook for an optional local Gemma assistant that can enhance the JSON before generation - disabled by default.

Inputs and outputs

Required: clip, design_data, ideoboard_input_signature. Optional: ideoboard. Outputs: positive (CONDITIONING - straight into a KSampler), prompt_json (the exact structured prompt sent to the model; paste it into a viewer to inspect), width, height (canvas size for an EmptyLatent or similar), and gemma_json_request (a STRING request payload for the Gemma assistant path).

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. You supply the Ideogram-class model and CLIP yourself. README floor: ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8.

Where people get burned

• It encodes with whatever CLIP you hand it. An Ideogram-style prompt fed into a CLIP that doesn't know the layout vocabulary will still produce an image - it just won't hold the grid or the text. Pair it with the right text encoder for your model.
• Text-in-frame only works if the model genuinely renders text; if yours doesn't, the panel text fields get ignored and you're paying JSON overhead for nothing. That's a model capability question, not a node bug.
• width/height are canvas outputs you must use to size your latent. Ignore them and the prompt and the empty latent can disagree about the sheet's proportions, which is a classic way to get panels that don't fit the frame.