ComfyUI Node
Difforum · H3 Guides
Anchor Difforum keyframes inside a MiniMax H3 generation. Wraps ComfyUI's core `MiniMaxH3AddGuide` once per keyframe, so the camera you drew becomes the H3 shot: feed keyframes + indices from the Keyframes node (grid H3) and the positive + AV latent from `MiniMax H3 Reference to Video` (or `Image to Video`). Optionally anchor a soundtrack at frame 0, so an audio-reactive direction and H3's own audio stay in sync. H3 is trained with a few guides per clip: `max_guides` keeps the first, the last and evenly spaced ones in between.
Difforum · H3 Guides
- positive
- latent
- vae
- keyframes
- audio_vae
- audio
- positive
- info
◄indices0►
◄max_guides4►
◄skip_firstfalse►
CategoryDifforum/5 · Video model bridges
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| positive | CONDITIONING | Positive conditioning from MiniMax H3 Reference to Video. | |
| latent | LATENT | The MiniMax H3 AV latent. | |
| vae | VAE | MiniMax H3 video VAE. | |
| keyframes | IMAGE | Keyframe pictures (Keyframes, Keyframe Images or Fill Reveal). | |
| indices | STRING | 0 | Frame number of each keyframe, comma separated, e.g. 0,48,96,123. |
| max_guides | INT | 41–16 | Most keyframes anchored inside the generation. The first and last are kept, the rest evenly spaced. More = H3 follows your frames closely; fewer = more of H3's own motion and invention. 4 for a camera path, 6-8 for a look pass. |
| skip_first | BOOLEAN | false | Enable when frame 0 is already set (e.g. Image to Video first_frame). |
| audio_vaeopt | VAE | MiniMax H3 audio VAE, needed with audio. | |
| audioopt | AUDIO | Soundtrack anchored at frame 0. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| positive | CONDITIONING | — |
| info | STRING | — |