How does the prompt cycling work?

Provide a multiline list of style prompts or let GeminiNode generate them using the Author Notes’ pattern. FL_PromptMulti assigns one prompt per frame and repeats the list after the final line, ensuring a predictable style cycle across the video.

Do I have to use GeminiNode for prompts?

No. You can paste hand-written prompts into PrimitiveStringMultiline. GeminiNode is optional and helps draft prompts; RegexReplace then normalizes them (e.g., removes numbering and enforces using “the source frame”).

Will the output have audio?

This setup focuses on frame-to-frame visual stylization. If your CreateVideo/SaveVideo configuration doesn’t pass through audio, you can remux the original audio onto the stylized video using a video editor or ffmpeg.

How can I reduce flicker between frames?

Keep prompts concise and consistent, always referencing “the source frame.” Avoid subject re-descriptions, keep style terms stable across frames, and process every frame (Nth=1) for best temporal consistency. Larger N values speed tests but increase perceptual jumps.

Stylize Video - Frame by Frame - Flux.2 Klein 4b

This workflow applies a consistent, frame-by-frame style transfer to any input video using the Flux.2 Klein 4b Distilled image editing model. It decodes your video with LoadVideo and reads resolution/FPS via GetVideoComponents, then uses VHS_SelectEveryNthImage to choose how densely to process frames (every frame, or every Nth for speed). Frames are organized with ImpactImageBatchToImageList so each image can be paired with a per-frame prompt.

Style prompts are generated and aligned with frames using a small prompt pipeline: GeminiNode can draft style-transfer lines following the Author Notes, RegexReplace cleans and normalizes the text (for example, removing numbering and enforcing "the source frame" phrasing), and FL_PromptMulti distributes one prompt per frame in a repeating cycle. Each frame and its corresponding prompt are fed into the Flux.2 Klein 4b editing node (the Flux.2 Klein 4b Distilled image editing model) to produce a stylized image while preserving the source frame’s structure. The processed images are repacked with ImageListToImageBatch, converted to a video using CreateVideo (matching the original FPS from GetVideoComponents), previewed via PreviewAny as needed, and written to disk with SaveVideo.

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.

41f029c74cd4.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