ComfyUI Node
Fantastic H3 Video Edit Latent
Turn a source clip and a mask into the latent MiniMax H3 samples for an edit: white areas of the mask are regenerated, everything else is held to the source. Use it in place of the empty latent, and wire width, height and length into the Text Encode. The Media Loader's Mask for editing does this for you through the RefMod Text Encode; this node is for masks made elsewhere.
Fantastic H3 Video Edit Latent
- vae
- images
- mask
- audio_vae
- audio
- latent
- width
- height
- length
◄grow16►
◄keep_audiotrue►
◄megapixels0.98►
Categoryconditioning/video_models
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| vae | VAE | The H3 video VAE. | |
| images | IMAGE | The source clip's frames, at 24 fps. | |
| mask | MASK | White where the clip is regenerated: whatever is being replaced, changed or removed. One mask per frame, or a single frame for the whole clip. | |
| grow | INT | 160–256 | Widen the mask by this many source pixels, so edges and shadows are replaced too. |
| keep_audio | BOOLEAN | true | On: the source audio is held unchanged. Off, or no audio connected: the soundtrack is generated fresh. |
| audio_vaeopt | VAE | The H3 audio VAE, needed to keep the source audio. | |
| audioopt | AUDIO | The source clip's soundtrack. | |
| megapixelsopt | FLOAT | 0.980–16 | Sample at no more than this many megapixels, keeping the clip's shape; never enlarged. 0 samples at the clip's own size. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | — |
| width | INT | — |
| height | INT | — |
| length | INT | — |