How should I structure the reference sheet for best consistency?

Place each character/prop/location clearly and non-overlapping on a neutral background, with consistent lighting and angles. Avoid heavy text overlays. Ensure each subject is large enough and well-isolated so the IC-LoRA can bind the visual identity. A single composite image is required—this workflow repeats the same sheet across all frames via RepeatImageBatch.

What does the split prompt do in this workflow?

IC-LoRA Ingredients expects a separation between the list of entities to anchor (ingredients) and the action narration. The ingredients list should reference the same names you intend from the sheet; the action describes motion, camera, or scene. Keeping them distinct helps the model preserve identity while animating. See the official IC-LoRA usage guide for exact formatting recommendations.

How do I tune video length and resolution without running out of VRAM?

Use the Video Settings group: set fps and seconds/frames with PrimitiveInt and let ComfyMathExpression compute total frames. Start with short clips and moderate fps (e.g., ~2–4 seconds). Choose a resolution that’s a multiple of 16 and use ResizeAndPadImage to match the EmptyImage canvas. If you hit VRAM limits, reduce resolution, lower fps, or shorten the clip.

How can I improve identity stability across frames?

Keep the ingredient list concise and consistent with the reference sheet, reduce overly complex action text, and avoid introducing new, unseen entities in the action. Increase IC-LoRA adherence (LoRA strength) in the LTX-2.3 node if available, ensure the reference sheet is high quality, and keep motion/camera changes moderate. Because RepeatImageBatch conditions every frame, stability improves when the reference is clean and unambiguous.

LTX-2.3 IC-LoRA: Reference Sheet Control

This workflow turns a composite reference sheet into a short, identity-consistent video using LTX-2.3 with the IC-LoRA Ingredients add-on. You provide a single image that inventories your characters, props, and/or locations, plus a split prompt where one part lists the "ingredients" from the sheet and the other describes the action. The core LTX-2.3 inference (custom node f9f61b10-b689-4d67-b4fa-0acc1d9b5390) uses the IC-LoRA to anchor visual identity to your sheet while animating the described motion.

Technically, the workflow loads your sheet with LoadImage, normalizes it with GetImageSize and ResizeAndPadImage, and then duplicates it across the timeline via RepeatImageBatch so every frame is conditioned on the same visual references. PrimitiveInt values in the Video Settings group (paired with ComfyMathExpression) control fps and total frames. An EmptyImage node defines the target canvas/resolution for generation, and PreviewImage lets you sanity-check inputs and a sample frame. Finally, SaveVideo encodes the generated frame batch into a video file. A MarkdownNote inside the graph includes quick usage tips and links to the official IC-LoRA documentation for prompt formatting and best practices.

API

Use this workflow from code

Every Comfy workflow is a JSON graph. The payload below is this workflow, exactly as ComfyUI runs it — fetch it from the URL, keep it in version control, or load it in ComfyUI and run it node by node.

82237f7d19b7.json

Fetching workflow JSON…

Run it from TypeScript or Python with the Comfy SDK. The same code targets Comfy Cloud or a ComfyUI you host yourself — only the base URL changes.

// Install (beta)
npm i @comfyorg/sdk

// Run this workflow (TypeScript)
import { Comfy } from "@comfyorg/sdk";

const client = new Comfy({ apiKey: "comfyui-..." });
const wf = await client.workflows.fromFile("workflow_api.json");
const job = await client.run(wf);
await job.getOutputs("<output-node-id>")[0].toFile("output.png");

The SDK takes a workflow in API format: open this workflow in ComfyUI and use File → Export Workflow (API).

Comfy Cloud API access requires a plan with an API key.

SDK docs Get an API key

FAQ

Frequently Asked Questions