Ideogram Caption Preview

Ideogram Studio is a visual editor, but sometimes your caption doesn't come from an editor. It comes out of an LLM node that turned "girl in red dress, head top-left" into proper Ideogram JSON. Or it's a saved caption you pasted in. Or it's an existing workflow's caption you're modifying programmatically. Ideogram Caption Preview is the node for that situation: give it a JSON caption string, get back the same extras bundle the studio produces - overlay image, alpha mask, and all - rendered from your caption alone.

Think of it as the studio's render engine without the drawing part. Same output format, no canvas.

The inputs that matter

• caption - the Ideogram JSON caption as a STRING (multiline, so paste away). This is the one you'll actually touch.
• width / height - the resolution the overlay is drawn at, default 1024×1024, 16–8192. Unlike the studio, the preview has no resolution chip, so you set these yourself. Match them to what the sampler will actually render or the overlay will lie about placement.
• line_width, fill_alpha, label_size - how the boxes are drawn: stroke width, fill opacity (0–1, default 0.18), and label font size. Cosmetic, but fiddly enough to matter when you're compositing over a dark image.
• show_index / show_text - toggles for the element number labels and the literal text on text boxes. Leave both on unless the overlay is getting cluttered.
• overrides - optional IDEOGRAM_OVERRIDE input, same patch mechanism as the studio. Handy for previewing a caption with a tweak applied, without editing the string.

The output

One output: extras, an IDEOGRAM_EXTRAS bundle - the exact thing the studio emits, and the exact thing Ideogram Studio Extras unpacks into overlay / alpha / width / height / bboxes. So the pattern is: Caption Preview → Extras → ImageCompositeMasked over your generation, and you've got a visual check of a caption you never drew by hand.

This also makes it the glue for the LLM-driven workflow that keeps showing up in the community: LLM emits JSON → Caption Preview renders it → you see the boxes before you spend a 3–4 minute generation on a malformed prompt. Catching layout errors in a second of overlay instead of after a render is a real time-saver with this model.

Installing it

It ships with the pack - no separate install:

cd ComfyUI/custom_nodes
git clone https://github.com/nynxz/ComfyUI-IdeogramHelper

Restart ComfyUI, find it under Ideogram → Ideogram Caption Preview. No pip dependencies; the only requirement is a ComfyUI recent enough for the V3 node API.

Where it bites

The node validates the caption and raises a clear error if the input isn't valid JSON - which is the feature, not the bug, but it means garbage in from your LLM fails loudly mid-graph. Keep the multiline caption field fed from a real JSON source, and remember the preview only renders; it doesn't re-serialize or clean the caption like the studio does. What you paste in is what gets drawn. The one trap worth knowing: fill_alpha and line_width interact - a fat line plus low alpha looks fine, but thin lines at 0.0 alpha are nearly invisible against a busy image.