MiniMax H3 Latent Upscale With Model - ComfyUI Built-in Node Documentation - ComfyUI

MiniMax H3 Latent Upscale With Model resizes the video part of a MiniMax H3 latent to a target width and height with a latent upscale model. The audio latent rides along untouched, so the clip keeps its sound, and no re-sampling is needed. Use it after MiniMax H3 sampling and before the H3 VAE decode when the clip should come out larger.

Inputs

Parameter	Description	Data Type	Required	Range
`model`	Latent upscale model, usually loaded with Load Latent Upscale Model. It has to be a MiniMax H3 3D-conv latent upscaler.	LATENT_UPSCALE_MODEL	Yes	-
`samples`	MiniMax H3 latent to resize. A nested latent keeps its audio part as it is.	LATENT	Yes	-
`width`	Target video width in pixels. The latent is resized to `width / 16` columns, so the value is rounded down to a multiple of 16. (default: 2560)	INT	Yes	32 to 16384 (step 32)
`height`	Target video height in pixels, also rounded down to a multiple of 16. (default: 1440)	INT	Yes	32 to 16384 (step 32)

When the requested size already matches the latent, the input latent is returned unchanged. A resized latent carries no noise_mask, so a partially masked latent loses its mask. The network runs its 3D convolutions in temporal slabs, which keeps memory use on long clips close to that of a single slab.

Outputs

Output Name	Description	Data Type
`LATENT`	The resized latent, ready for the H3 VAE decode.	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): cd659baace5a21620033dd3470b854cccbada01d09545a2f61273b949c3a004b