ComfyUI Node
H3 Latent Upscale Pixaroma
Makes a MiniMax H3 video latent bigger with a trained latent upscaler, before H3 re-draws it at the new size in a few steps (a 2K regenerate pass). Wire the latent of your clip in (VAE Encode with the H3 video VAE, or an H3 sampler's output), set the new width and height, and send the result on to H3 Audio Sync or the sampler. Picture and sound stay together; a latent with no sound part gets a silent one. Needs an H3 latent upscaler file in models/latent_upscale_models. Based on the MiniMax H3 latent upscaler by LBH-123-AI (MIT); the math is unchanged, so the result matches the original node.
H3 Latent Upscale Pixaroma
- latent
- latent
◄upscale_model▾►
◄width1440►
◄height2560►
Category👑 Pixaroma/✂️ Resize & Crop
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| latent | LATENT | The MiniMax H3 latent to make bigger: VAE Encode with the H3 video VAE, or the latent out of an H3 sampler. Picture and sound both work. | |
| upscale_model | COMBO | The H3 latent upscaler file, from models/latent_upscale_models. minimax_h3_latent_upscaler_3d_conv_v1 is the one NVIDIA's 2K regenerate uses. | |
| width | INT | 144064–8192 | The new width in picture pixels. Rounded to a multiple of 32, which H3 needs. Keep the shape of your video: 1440 x 2560 is 2K portrait, 2560 x 1440 2K landscape. |
| height | INT | 256064–8192 | The new height in picture pixels. Rounded to a multiple of 32, which H3 needs. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | The bigger H3 latent, picture and sound together. Send it to H3 Audio Sync (to keep your own sound) or to the sampler that re-draws it. |