MiniMax H3 · ComfyUI

Edit one frame.
Ripple the rest.

Make your change in an image editor. Ripple carries it through the video, using your edited frames as anchors.

Swap a subject, rework a scene, or change a product. Keep your own editing tools—or run the image edits locally with Qwen.

The full demo

See Ripple in action.

1:17 · With audio

See the change

Same shot. New possibilities.

Original leftRipple right

01 / Environment

From green screen to main stage

A green-screen performance becomes a concert-stage scene.

Open video

Actual output comparisons. Clips are silent; use the player controls to watch or view fullscreen.

The idea is simple

An image edit becomes a video edit.

  1. 01

    Choose your moment

    Load a video, choose the shot, and place anchor markers on the frames you want to edit. Adjust the crop and output size in the same view.

  2. 02

    Make the change

    Edit the marked frames in your preferred image editor, or connect a local model. Add more anchors when other views need the same change.

  3. 03

    Let it ripple

    H3 uses the source video and edited anchors to propagate the change through the shot. Save the result and a side-by-side comparison.

Make it yours

Two ways to work.

Download a workflow JSON.
Drag it onto your ComfyUI canvas.

Bring your own editor01

Ripple Simple

Edit anchor frames with the tools you already know. Download a frame, make your change, and upload the edited image to its matching marker.

  • Your choice of image editor
  • Five local model files
  • Model links included in the workflow
Download Simple

H3_Ripple_Simple.json

Edit inside ComfyUI02

Ripple Local Editor

Use two Qwen Image 2.1 branches to edit your anchors locally, then pass those images directly into Ripple. Adapt the graph to your own workflow.

  • Two connected image-editing branches
  • Eight local model files
  • No API required for the local path
Download Local Editor

H3_Ripple_Local_Editor.json

Bring your own source video. Both downloads include a Models & setup card and are configured for the H3 Turbo 4-step LoRA.

Start here

A little setup.
Then your shot.

You’ll need a current ComfyUI installation and hardware capable of running MiniMax H3. The local editor also runs Qwen Image 2.1.

Full installation guide
01

Install the custom nodes

Clone ComfyUI-WepeNerd into ComfyUI/custom_nodes, install its requirements with ComfyUI’s Python environment, then restart ComfyUI.

02

Get the models

Open the workflow’s Models & setup card for exact downloads and folders. Both workflows need H3 FL2VA, its text encoder and video VAE, the Turbo LoRA, and the Ripple LoRA.

Local Editor adds Qwen Image 2.1, its text encoder and VAE. Both image-editing branches share those three files.

03

Load your video and edit

Select your own video in Ripple Clip. Choose the shot, output size and anchors. In Simple, upload your edited frame; in Local Editor, replace both example edit instructions with your intended change.

Ripple Output is enabled in these downloads, so queuing also renders the H3 video. Start with a small output size and keep your anchor edits consistent.

Model weightsH3 Ripple on Hugging Face Code & documentationComfyUI-WepeNerd on GitHub