CosmosImageToVideoLatent - ComfyUI Built-in Node Documentation

The CosmosImageToVideoLatent node creates a video latent representation from input images. It builds a blank video latent with the requested width, height, and number of frames, then optionally encodes a start image into the beginning frames and/or an end image into the end frames. When images are provided, it also generates a noise mask so the encoded frames remain fixed during generation.

Inputs

Parameter	Description	Data Type	Required	Range
`vae`	The VAE model used to encode the images into the video latent space.	VAE	Yes	-
`width`	The width of the output video in pixels (default: 1280).	INT	Yes	16 to MAX_RESOLUTION (step: 16)
`height`	The height of the output video in pixels (default: 704).	INT	Yes	16 to MAX_RESOLUTION (step: 16)
`length`	The total number of frames in the video (default: 121).	INT	Yes	1 to MAX_RESOLUTION (step: 8)
`batch_size`	The number of video latents to generate (default: 1).	INT	Yes	1 to 4096
`start_image`	Optional image or image sequence to encode at the beginning of the video.	IMAGE	No	-
`end_image`	Optional image or image sequence to encode at the end of the video.	IMAGE	No	-

Note:

When neither start_image nor end_image is provided, the node returns a blank latent without a noise mask.
When start_image is provided, it is encoded into the first frames of the latent and those frames are marked with noise mask value 0 (preserved). When end_image is provided, it is encoded into the last frames and those frames are marked with noise mask value 0. The remaining frames keep a mask value of 1.
The latent has 16 channels and its spatial dimensions are height / 8 by width / 8. The number of latent frames is ((length - 1) // 8) + 1.
batch_size repeats the latent and, when present, the noise mask.

Outputs

Output Name	Description	Data Type
`latent`	The generated video latent containing the optional encoded start and/or end images and, when images are provided, a matching noise mask with value 0 on the preserved frames.	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): 0b06ccfcb14c27c81eeebbbff519da1e187970d4cfc19c8796fc3da20688245c

​Inputs

​Outputs

Inputs

Outputs