How does this workflow keep motion stable while changing the style?

It computes a per-frame depth map from your input video to preserve scene structure, then uses ByteDance2ReferenceNode (Seedance 2.0) to blend that structure with a style reference created from the first frame by GPT-Image-2. Increasing depth/consistency settings (if exposed) typically reduces flicker and warping.

Do I need an OpenAI API key for GPT-Image-2?

Yes. OpenAIGPTImageNodeV2 calls GPT-Image-2 to create a style-matched reference from the first frame. Add your API key in that node’s settings and adjust any available parameters to control the reference output.

Can I generate multiple styled versions at once?

Yes. Enter up to four prompts (one per line) in PrimitiveStringMultiline. The batch nodes (a3edb32a-4533-45e3-8eb2-84399225da2e) fan out the inputs so the pipeline produces up to four videos in one run.

Why is my result flickering or drifting from the source motion?

Try increasing the depth influence or temporal consistency in ByteDance2ReferenceNode, lowering the style strength, and using a fixed seed. Also ensure the input fps and resolution are sensible for your GPU; excessively high settings can cause instability.

Does the output video include audio?

No. SaveVideo writes silent video files. If you need audio, export it from the original clip and remux it onto the styled output in your video editor.

Style Transfer Vid2Vid - GPT Image 2 + Seedance 2.0

This depth-aware, reference-guided video-to-video workflow applies consistent style transfer to your footage while preserving motion and scene geometry. It starts by ingesting your clip with LoadVideo, then computes a per-frame depth map to anchor structure and camera movement (handled by the custom processing node c3b10638-32f0-4101-9b21-21fcda8712d8). The very first frame is extracted and saved (via SaveImage and helper utilities like 8dbec705-cae4-4197-9594-85b48cc18483), then passed to OpenAIGPTImageNodeV2 together with your text prompt from PrimitiveStringMultiline. GPT-Image-2 generates a style-matched reference image from that initial frame, which the workflow uses to steer the overall look.

The core stylization is driven by Seedance 2.0 through ByteDance2ReferenceNode, which fuses three signals: the depth map (structure), the first-frame style reference (appearance), and your prompt (semantic guidance). You can queue up to four style variations in one run using the batch fan-out/collector nodes (a3edb32a-4533-45e3-8eb2-84399225da2e), then reassemble the generated frames into final videos with CreateVideo and SaveVideo. The result is a set of stable, on-style clips that maintain the original timing and motion while reflecting your chosen aesthetic.

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.

e1e03cafda18.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