Wan22ImageToVideoLatent - ComfyUI Built-in Node Documentation

Wan22ImageToVideoLatent creates video latent representations from images. It generates a blank video latent space with the specified width, height, frame length, and batch size, and can optionally encode a starting image sequence into the beginning frames. When a start image is provided, the node encodes it into the latent space and creates a corresponding noise mask that marks which regions should be denoised during generation.

Inputs

Parameter	Description	Data Type	Required	Range
`vae`	The VAE model used to encode the start image into latent space	VAE	Yes	-
`width`	The width of the output video in pixels (default: 1280, step: 32)	INT	Yes	32 to MAX_RESOLUTION
`height`	The height of the output video in pixels (default: 704, step: 32)	INT	Yes	32 to MAX_RESOLUTION
`length`	The number of frames in the video sequence (default: 49, step: 4)	INT	Yes	1 to MAX_RESOLUTION
`batch_size`	The number of video latents to generate (default: 1)	INT	Yes	1 to 4096
`start_image`	Optional starting image sequence to encode into the beginning frames of the video latent (uses the first `length` frames)	IMAGE	No	-

Note: When start_image is provided, the image sequence is upscaled to the target width and height, encoded with the VAE, and placed into the first frames of the latent. The noise mask for those frames is set to 0 (preserved), while the remaining frames have a mask value of 1 (to be denoised). The latent always has 48 channels, spatial dimensions of height / 16 by width / 16, and a temporal dimension of ((length - 1) // 4) + 1. width and height must be divisible by 16 (enforced by the step of 32), and length increases the temporal dimension in steps of 4.

Outputs

Output Name	Description	Data Type
`samples`	The generated video latent representation, repeated for each item in the batch	LATENT
`noise_mask`	The noise mask indicating which regions should be denoised (value 1) and which retain the encoded start image (value 0)	LATENT

Both fields are returned together inside a single LATENT output.

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): 3d05980641eeef2e86df7a845aa8b2bd703882db98fe71adef2746ab34a9d717

​Inputs

​Outputs

Inputs

Outputs