Skip to main content

Wan 2.2 Animate · Advanced ComfyUI workflow

Your character.
The performance you choose.

Start with a full-body character image and a video of the movement you want. This workflow uses the performance to animate the character, so you can direct the motion with a clip instead of describing every pose.

Wan 2.2 Animate · 4.8 sec
Character imageMotion videoGenerated result
The image sets the character; the video guides the movement.The output follows the dance; clothing and small mechanical details can change.Open comparison video

New inputs need subject-mask setup in ComfyUI. Review the GPU rate before launching: startup and idle time are paid. Download your work, then remove the machine in ModelPilot; stopping it can leave paid storage.

From your inputs to a take you can review

What you supply. What ModelPilot handles.

  1. 01 · Your inputs

    Pick the character and performance

    Use a full-body image and a motion video with visible limbs. Similar silhouettes give the workflow less to reconcile. Use inputs you have permission to use.

  2. 02 · Hosted workspace

    Open ComfyUI with the models installed

    ModelPilot provisions the GPU and model environment. To match this example, download the short V4.1 workflow and drag it onto ComfyUI; the workspace initially opens a longer graph. Upload your inputs, preview the subject mask and adjust the SAM2 mask points where it selects the wrong area.

  3. 03 · Your controls

    Render, inspect and change the take

    Keep the graph editable: change inputs, masks or generation settings, then download the takes you want. Check hands, clothing and character details before using the result.

Useful for trying a performance on a character.

Motion you can compare

The supplied dance and output play side by side. The 4.8-second example follows the dance, including turns and arm movements.

Details still need inspection

Hip panels and footwear vary even in this short result. If exact costume continuity matters to your shot, this example does not establish that it will meet your needs.

Best for hands-on ComfyUI users

The mask needs visual adjustment for new inputs. Allow time for setup and iteration; there is no guaranteed first-render match.

Longer example: where the character starts to change

This 19.0625-second V4.2 take has 305 frames across four generation windows. The dance continues, but footwear, hip panels and the back of the head change as it runs. Longer duration is not evidence of better character consistency.

The successful mask check and render used about 25 minutes and $0.20 in the July 2026 experiment. Total experiment spend was about $0.41, including a rerun after a download failure. These are historical observations, not a quote for your session.

Workflow versions, settings and source credits

The download contains motion-transfer-v4.1.json for the 77-frame short configuration and motion-transfer-v4.2-long-305f.json for the longer example, plus pinned model and node versions and mask instructions.

The hosted workspace opens a longer workflow. To use the short configuration, download the V4.1 JSON from the Gist and drag it onto the ComfyUI canvas. Replace the referenced inputs and check the mask before running it.

  • Short result: 77 frames, 16 fps, 4.8125 seconds. The side-by-side comparison is 1728 × 1024.
  • Driver: adapted from the robot dance by AdisResic on Pixabay; the short source interval is 14.0–18.8125 seconds.
  • The fitted character image was made with image generation. It is a reference input, not a photograph of a finished animation.

Driver by AdisResic · Pixabay Content License

Download workflows and setup notes