ComfyUI Node
SceneCut Detect
Find every cut in a video with OmniShotCut, clean the result up with a preset and check each cut with DINOv3. Models download on first use (OmniShotCut 164 MB; DINOv3 1.2 GB and SSCD 178 MB when validating).
SceneCut Detect
- video
- scenes
- shot_count
- summary
- images
◄presetFilm / TV►
◄validatetrue►
◄devicedefault►
◄output_imagesoff►
CategorySceneCut
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| video | VIDEO | — | |
| preset | COMBO | Film / TV | Film / TV: Measured pacing with occasional dissolves and fades. Shots under 0.4 s are treated as detector noise. Music video / fast edit: Cuts on the beat, very short shots, strobes and flashes. Keeps shots down to ~0.17 s and treats more light changes as flashes rather than cuts. Anime / animation: Held drawings, impact flashes, smear frames and frequent fades to black. Short shots kept down to 0.25 s; flashes detected aggressively. Vlog / talking head: Jump cuts inside one setup and few real scene changes. Shots under 0.5 s are merged. Sports / broadcast: Fast pans, graphic wipes into replays, logo stings. Transitions are kept as their own segments so replays can be found. |
| validate | BOOLEAN | true | Check every cut with DINOv3 and merge false ones (flashes, fast motion, a continuous shot split in two). Slower; off uses the detector's cuts as refined. |
| device | COMBO | default | default: ComfyUI's GPU, managed with the other models. cpu: slow but needs no GPU. |
| output_images | COMBO | off | Also send frames to the images output, from the selected scenes (all scenes when none are selected). all frames: every frame, in order. first / middle frame per scene: one image per scene. off: the images output carries a single black frame. Frames are full resolution float32 (about 25 MB per 1080p frame), so prefer one per scene or select scenes first. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| scenes | SCENECUT_SCENES | — |
| shot_count | INT | — |
| summary | STRING | — |
| images | IMAGE | Frames chosen by output_images (a single black frame when off). |