Skip to main content
This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub
The CosmosPredict2ImageToVideoLatent node creates video latent representations from images for video generation. It can generate a blank video latent or incorporate start and end images to create video sequences with specified dimensions and duration. The node handles the encoding of images into the appropriate latent space format for video processing.

Inputs

ParameterData TypeRequiredRangeDescription
vaeVAEYes-The VAE model used for encoding images into latent space
widthINTNo16 to MAX_RESOLUTIONThe width of the output video in pixels (default: 848, must be divisible by 16)
heightINTNo16 to MAX_RESOLUTIONThe height of the output video in pixels (default: 480, must be divisible by 16)
lengthINTNo1 to MAX_RESOLUTIONThe number of frames in the video sequence (default: 93, step: 4)
batch_sizeINTNo1 to 4096The number of video sequences to generate (default: 1)
start_imageIMAGENo-Optional starting image for the video sequence
end_imageIMAGENo-Optional ending image for the video sequence
Note: When neither start_image nor end_image are provided, the node generates a blank video latent. When images are provided, they are encoded and positioned at the beginning and/or end of the video sequence with appropriate masking.

Outputs

Output NameData TypeDescription
samplesLATENTThe generated video latent representation containing the encoded video sequence
noise_maskLATENTA mask indicating which parts of the latent should be preserved during generation