Video to video

Making a new clip that takes its motion, layout or camera from an existing video.

In plain words

A source video can guide a new clip in several ways:

Restyle by partial denoise: encode the video into a latent and redraw it with denoise below 1, as image to image does for pictures. Motion stays, the look changes; too low a denoise changes little, too high loses the motion.
Control signals: turn each frame into a depth map, a pose skeleton, edges or line art, and let the model follow them while the prompt sets the look. Wan uses VACE or Wan Fun Control models for this.
Pose transfer onto a character from a picture: see Pose animation.
Camera re-shoot: ReCamMaster redraws the scene along another camera path.

Load Video (file from input) → Get Video Components gives images (frames), audio and fps.
Preprocessors from the comfyui_controlnet_aux pack (Depth Anything, DWPose, OpenPose, Canny, Lineart) between the frames and the control input.
WanVideoWrapper: WanVideo VACE Encode (input_frames, ref_images, strength, start and end percent); built-in ComfyUI has WanVaceToVideo for the same idea.
Resize and batch-cut nodes (ImageScale, ImageFromBatch) to match the clip's size and length.

Control strength about 1; lower for more freedom.
Ending the control early (vace_end_percent 0.5) keeps the layout but lets the model finish the detail.
The new clip cannot be longer than the source frames you give.

Source frames of another size or length than the clip.
Raw frames when you wanted only the motion: the colours and clothes come along. Use depth or pose.
Using someone's video without the right to: see Content rules.