WanAnimate2ToVideo - ComfyUI Built-in Node Documentation - ComfyUI

WanAnimate2ToVideo animates a character from a reference image by transferring the facial expressions, body motion, and hand gestures from a separate pose video. It builds the conditioning data and a starting latent that a video generation sampler uses to create the animation.

Inputs

Parameter	Description	Data Type	Required	Range
`positive`	The positive conditioning for the video generation.	CONDITIONING	Yes	N/A
`negative`	The negative conditioning for the video generation.	CONDITIONING	Yes	N/A
`vae`	The VAE used to encode the reference image and video frames into latent space.	VAE	Yes	N/A
`width`	Output video width in pixels. (default: 832)	INT	Yes	16 to MAX_RESOLUTION (step 16)
`height`	Output video height in pixels. (default: 480)	INT	Yes	16 to MAX_RESOLUTION (step 16)
`length`	Number of frames to generate. (default: 81)	INT	Yes	1 to MAX_RESOLUTION (step 4)
`batch_size`	Number of videos to generate simultaneously. (default: 1)	INT	Yes	1 to 4096
`reference_image`	The character to animate. If omitted, a black image is used.	IMAGE	No	N/A
`pose_video`	The video whose motion is transferred to the reference character. If it has fewer frames than `length`, the last frame is repeated to fill the missing frames.	IMAGE	No	N/A
`clip_vision_output`	CLIP vision of the reference image.	CLIP_VISION_OUTPUT	No	N/A
`positive_pose`	Prompt for the pose-video branch, describing the motion rather than the character. Defaults to `positive`. Used for both the cond and uncond passes.	CONDITIONING	No	N/A
`clip_vision_output_pose`	CLIP vision of the pose video’s first frame. Defaults to `clip_vision_output`.	CLIP_VISION_OUTPUT	No	N/A
`continue_motion`	Previous motion sequence to continue from for temporal consistency. Only the last frame of this sequence is used as the starting motion frame.	IMAGE	No	N/A
`video_frame_offset`	Frames to seek into the pose video. Connect to the `video_frame_offset` output of the previous node when extending. (default: 0)	INT	Yes	0 to MAX_RESOLUTION
`pose_strength`	Scales the pose video’s influence on the motion. 1.0 is the trained behavior; below weakens adherence, above amplifies. 0.0 mutes it but does not fully remove it. (default: 1.0)	FLOAT	Yes	0.00 to 10.00 (step 0.01)
`pose_start_percent`	Sampling percent at which the pose influence starts. Outside the window the pose branch is skipped entirely, which also speeds those steps up. (default: 0.0)	FLOAT	Yes	0.00 to 1.00 (step 0.01)
`pose_end_percent`	Sampling percent at which the pose influence ends. Motion is mostly established early, so e.g. 0.7 can loosen fine detail while keeping the choreography. (default: 1.0)	FLOAT	Yes	0.00 to 1.00 (step 0.01)
`reference_image_strength`	Scales how strongly generated frames attend to the reference image’s latent frame. Below 1.0 loosens identity/appearance adherence (e.g. to let the prompt restyle), above tightens it against drift. (default: 1.0)	FLOAT	Yes	0.00 to 10.00 (step 0.01)

Validation notes:

pose_start_percent must not be greater than pose_end_percent; otherwise the node raises a ValueError.
If pose_video is provided, its frame count must be greater than video_frame_offset; otherwise the node raises a ValueError.

Outputs

Output Name	Description	Data Type
`positive`	Positive conditioning for sampling, with the reference image, mask, and optional pose data attached.	CONDITIONING
`negative`	Negative conditioning for sampling, with the same reference image, mask, and optional pose data attached.	CONDITIONING
`latent`	Zero-filled starting latent for the video sampler; the first `trim_latent` frames should be removed before decoding.	LATENT
`trim_latent`	Number of latent frames that should be trimmed before decoding.	INT
`trim_image`	Number of overlapping image frames when extending a video.	INT
`video_frame_offset`	Frames to seek into the pose video; equals the adjusted input offset plus the number of frames generated.	INT

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

Source fingerprint (SHA-256): 7e1f497983ab63a68e5ef5439b3ef4e9295f79f78530c9dc5de16a8238475f05