ComfyUI Node
MiniMax H3 Inpaint Latent (plate + mask)
Encode a plate into H3's AV latent and attach a mask so the sampler inpaints only the masked region (insert the ref subject). Wire the LATENT into SamplerCustomAdvanced's latent_image, and keep MiniMaxH3ReferenceToVideo for the conditioning.
MiniMax H3 Inpaint Latent (plate + mask)
- vae
- plate
- mask
- LATENT
◄width1024►
◄height1024►
◄length124►
◄invert_maskfalse►
Categorymodel/latent/minimax
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| vae | VAE | H3 video VAE (minimax_h3_video_vae) | |
| plate | IMAGE | Plate/background video frames | |
| mask | MASK | White = region to regenerate/insert; black = keep the plate | |
| width | INT | 102432–8192 | — |
| height | INT | 102432–8192 | — |
| length | INT | 1245–3600 | Frame count @24fps, snapped to the 17k+5 grid |
| invert_mask | BOOLEAN | false | Invert the mask if your white/black regions are swapped |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| LATENT | LATENT | — |