ComfyUI Node
Minimax H3 Latent Upscaler (3D)
A ComfyUI node in video/MinimaxH3 with 8 inputs and 1 output.
Minimax H3 Latent Upscaler (3D)
- latent
- latent
◄model_name▾►
◄mode▾►
◄align32►
◄keep_proportiontrue►
◄devicecuda►
◄precisionfp16►
◄offload_after_upscalefalse►
Categoryvideo/MinimaxH3
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| latent | * | Input latent (image or video). | |
| model_name | COMBO | Minimax H3 upscale model. | |
| mode | COMBO | How the target size is computed. | |
| align | INT | 321–512 | Pixel-space alignment. Both output axes are placed on a grid compatible with this value and the H3 16x VAE grid. |
| keep_proportion | BOOLEAN | true | Preserve the source aspect ratio as closely as the requested alignment grid allows. |
| device | COMBO | cuda | 2 options: cuda, cpu |
| precision | COMBO | fp16 | 3 options: fp32, fp16, bf16 |
| offload_after_upscale | BOOLEAN | false | Move the cached learned upscaler to CPU after inference to free VRAM. Leave off for faster repeated runs when VRAM is available. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| latent | * | Upscaled latent. |