WanVaceToVideo - ComfyUI Built-in Node Documentation

The WanVaceToVideo node prepares video conditioning data for video generation models. It takes positive and negative conditioning inputs along with an optional control video, masks, and reference image, and encodes them into latent representations that guide video generation. The node handles upscaling, padding, masking, and VAE encoding to build the appropriate conditioning structure for video models.

Inputs

Parameter	Description	Data Type	Required	Range
`positive`	Positive conditioning input for guiding the generation	CONDITIONING	Yes	-
`negative`	Negative conditioning input for guiding the generation	CONDITIONING	Yes	-
`vae`	VAE model used for encoding images and video frames	VAE	Yes	-
`width`	Output video width in pixels (default: 832, step: 16)	INT	Yes	16 to MAX_RESOLUTION
`height`	Output video height in pixels (default: 480, step: 16)	INT	Yes	16 to MAX_RESOLUTION
`length`	Number of frames in the video (default: 81, step: 4)	INT	Yes	1 to MAX_RESOLUTION
`batch_size`	Number of videos to generate simultaneously (default: 1)	INT	Yes	1 to 4096
`strength`	Condition strength for VACE control (default: 1.0, step: 0.01). This is not a LoRA strength. LoRA weights are applied through separate LoRA nodes.	FLOAT	Yes	0.0 to 1000.0
`control_video`	Optional input video used for control conditioning. If not provided, a neutral gray video is created automatically.	IMAGE	No	-
`control_masks`	Optional masks that determine which parts of the control video are active. If not provided, a full white mask is used.	MASK	No	-
`reference_image`	Optional reference image for additional conditioning. When provided, it is encoded and prepended to the latent sequence.	IMAGE	No	-

Note: When control_video is provided, it is truncated to length frames and upscaled to the specified width and height; if it has fewer frames than length, the missing frames are padded with neutral gray (value 0.5). When it is not provided, a neutral gray video of length frames is created automatically. control_masks are upscaled to the specified width and height, truncated to length frames, and padded with value 1.0 if shorter. The mask separates the control video into inactive and reactive parts, each VAE-encoded and concatenated along the channel dimension; the mask is also downsampled to latent resolution. When reference_image is provided, it is VAE-encoded and prepended to the latent sequence. The latent frame count is calculated as ((length - 1) // 4) + 1, and the latent spatial dimensions are height / 8 and width / 8.

Outputs

Output Name	Description	Data Type
`positive`	Positive conditioning with video control data (vace_frames, vace_mask, vace_strength) applied	CONDITIONING
`negative`	Negative conditioning with video control data (vace_frames, vace_mask, vace_strength) applied	CONDITIONING
`latent`	Empty latent tensor ready for video generation with shape [batch_size, 16, latent_length, height/8, width/8]	LATENT
`trim_latent`	Number of latent frames to trim when a reference image is used; 0 if no reference image is provided	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): 2039b7509ce5b731e9e41d9cd2dad022d4c5004751f571a4cf88c1ba0cae405b

​Inputs

​Outputs

Inputs

Outputs