LTXVLatentUpsampler - ComfyUI Built-in Node Documentation - ComfyUI

The LTXVLatentUpsampler node increases the spatial resolution of a video latent representation by a factor of two. It uses a specialized upscale model to process the latent data, which is first un-normalized and then re-normalized using the provided VAE’s channel statistics. This node is designed for video workflows within the latent space.

Inputs

Parameter	Description	Data Type	Required	Range
`samples`	The input latent representation of the video to be upscaled.	LATENT	Yes
`upscale_model`	The loaded model used to perform the 2x upscaling on the latent data.	LATENT_UPSCALE_MODEL	Yes
`vae`	The VAE model used to un-normalize the input latents before upscaling and to normalize the output latents afterwards.	VAE	Yes

Note: This node is marked as experimental in ComfyUI.

Outputs

Output Name	Description	Data Type
`LATENT`	The upscaled latent representation, with spatial dimensions doubled compared to the input. The output latent has the same batch size, number of channels, and temporal length as the input. The `noise_mask` from the input, if present, is removed from the output.	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): 7d7f0b733cb3758e9ec985cac30134d719b130b5b86c35bfdd14576a5b4575db