ComfyUI Node
Mpi H3 Pin First Frame
Pin a MiniMax H3 clip's first frame: writes one encoded frame into latent step 0 and masks that step out of sampling, video only - the audio stays fully generated. Step 0 is the VAE's 1-frame step, so one image is exactly one step and the pin is phase-exact; nothing to trim. Reference-to-video sets no keyframe, so the same seed, length and canvas still open on a different framing per prompt; clips that must join (Infinite Video roots) share a framing only when they share frame 0.
Mpi H3 Pin First Frame
- latent
- first_frame
- latent
CategoryMpiNodes/Utils
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| latent | LATENT | The TARGET AV latent - straight from MiniMaxH3ReferenceToVideo, MiniMaxH3ImageToVideo or EmptyMiniMaxH3LatentAV. A noise mask already on it (MpiH3MaskedPrefix) is kept. | |
| first_frame | LATENT | The frame to start on, encoded: one image through the video VAE (1 latent step), or any H3 latent - video-only or AV - whose FIRST frame is the one wanted (e.g. a sibling clip's saved latent). Same canvas. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | — |