WanImageToVideo - ComfyUI Built-in Node Documentation

The WanImageToVideo node prepares conditioning and latent representations for video generation. It creates an empty latent space for the video and can optionally incorporate a starting image and CLIP vision output to guide the generation. Both the positive and negative conditioning inputs are updated with the provided image and vision data.

Inputs

Parameter	Description	Data Type	Required	Range
`positive`	Positive conditioning input used to guide the generation	CONDITIONING	Yes	-
`negative`	Negative conditioning input used to guide the generation	CONDITIONING	Yes	-
`vae`	VAE model used to encode images into latent space	VAE	Yes	-
`width`	Width of the generated video (default: 832, step: 16)	INT	Yes	16 to MAX_RESOLUTION
`height`	Height of the generated video (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 in one batch (default: 1)	INT	Yes	1 to 4096
`clip_vision_output`	Optional CLIP vision output added as extra conditioning to both the positive and negative inputs	CLIP_VISION_OUTPUT	No	-
`start_image`	Optional starting image used to initialize the video. When provided, it is resized to the specified `width` and `height` and placed at the beginning of the frame sequence; any frames beyond `length` are ignored. The remaining frames are filled with neutral gray (0.5) values unless `ref_pad_image` is supplied.	IMAGE	No	-
`ref_pad_image`	Optional reference image whose first frame replaces the neutral gray padding of the frame sequence. It is resized to the specified `width` and `height`, and only the first image of the batch is used. Has an effect only when `start_image` is also provided.	IMAGE	No	-

Note: When start_image is provided, the frame sequence is encoded with the VAE and a mask is applied to the conditioning. The mask is set to 0 for the frames covered by the starting image and 1 for the remaining frames, so generation continues from the provided image. Only the first three color channels (RGB) of the image are used during encoding. Both positive and negative conditioning receive the same concatenated latent image, mask, and (if supplied) CLIP vision output. When ref_pad_image is supplied together with start_image, its first frame is resized to width and height and written into the RGB channels of the padding frames before the starting image is placed on top, so the padding carries the reference image instead of flat gray. This is SVI-style anti-drift padding, used by models such as ID-V2V.

Outputs

Output Name	Description	Data Type
`positive`	Positive conditioning, updated with the image and vision data	CONDITIONING
`negative`	Negative conditioning, updated with the image and vision data	CONDITIONING
`latent`	Empty latent tensor ready for video generation, with shape [batch_size, 16, ((length-1)//4)+1, height//8, width//8]	LATENT

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): 3000c1c816d2c123fc5bc46ea1f193c52c0f81a2f8c4a9110d8a4fa909185aea

​Inputs

​Outputs

Inputs

Outputs