ComfyUI Node
Difforum · Keyframes
Pick keyframes out of any frame batch (a Storyboard, a Feedback render, guide frames) for a video model, on that model's frame grid. Outputs the keyframes, their indices, a full-length *sparse* batch (black except at keyframes - what `LTXVAddGuidesFromBatch` expects), the first and last frame (for first-last-frame models) and the snapped clip length.
Difforum · Keyframes
- frames
- masks
- keyframes
- indices
- sparse_batch
- first_frame
- last_frame
- length
- key_masks
- info
◄gridLTX-2 / 2.5 (8k+1)►
◄every_seconds1.00►
◄fps24.00►
◄indices►
CategoryDifforum/5 · Video model bridges
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| frames | IMAGE | The clip to pick keyframes from (Guide Frames, Storyboard, a Feedback render...). Keyframes carry its look into H3 / LTX: better frames here = better video. | |
| grid | COMBO | LTX-2 / 2.5 (8k+1) | Snap keyframes to the video model's latent grid (H3 17k+5, LTX 8k+1, Wan 4k+1) so each one lands on a real latent frame. |
| every_seconds | FLOAT | 1.000–60 | How many keyframes: one every N seconds. 1 = many, strong guidance; 2 = fewer, more freedom; 0 = first and last only. H3 Guides then keeps up to max_guides. |
| fps | FLOAT | 24.001–120 | Frames per second of the clip (from Setup). |
| indicesopt | STRING | Exact frames instead of spacing, e.g. '0, 48, 96, -1' (-1 = last). | |
| masksopt | MASK | Guide Frames masks, so Fill Reveal can repaint what the camera uncovered. |
Outputs (8)
| Name | Type | Description |
|---|---|---|
| keyframes | IMAGE | — |
| indices | STRING | — |
| sparse_batch | IMAGE | — |
| first_frame | IMAGE | — |
| last_frame | IMAGE | — |
| length | INT | — |
| key_masks | MASK | — |
| info | STRING | — |