Every cut in your video, found, checked, fixed and exported, in ComfyUI.

Live demo

This is the SceneCut Viewer, the same code that runs inside the node, on the trailer of William Castle's Let's Kill Uncle (1966, public domain): hard cuts, a jump cut, dissolves, a fade, title cards and green-screen split screens. SceneCut found 23 scenes with the Film preset. The lower track shows the 4 boundaries it merged: a camera move and an animated graphic that validation judged to be one shot, a title too short to be a shot, and two transitions folded into the shot before. Scene 3 begins after a dissolve; click it to see why validation kept that cut.

What it does

Simple nodes with presets do the whole job; advanced nodes split the same pipeline into stages and expose every parameter.

Detect every shot

OmniShotCut predicts whole shots and labels transitions (dissolve, wipe, push, slide, zoom, fade, doorway) and jump cuts. Presets for film, music videos, anime, vlogs and sports, in seconds, so they mean the same at any frame rate.

Check every cut

DINOv3 compares the frames on both sides of each cut; camera flashes, strobes, whip pans and a shot split in two are merged back. Every decision keeps its scores, and nothing is thrown away.

Fix a cut in one click

Put the playhead on a frame on the wrong side of a cut and press Fix cut: the frames around it are compared with both scenes and the cut moves to where the picture really changes. Or join the frame to the nearest cut by hand.

Pick your scenes

Select scenes in the viewer and play them one at a time in Selected Scenes, in the node or in its own browser tab, kept in sync. Fixes and selection are saved with the workflow.

Export clips

One file per scene, cut on the exact frames, with its sound, plus scenes.json: from the viewer's Export clips button or the Save Scenes node. H.264 by default; H.265, NVENC, ProRes and lossless FFV1 too.

Plug into any workflow

Every node that outputs scenes can output their frames as an IMAGE batch (all, or one per scene), and the viewers take images in. Scene lists are JSON, and the core runs without ComfyUI.

Get started

Three steps to your first scene list. You need ComfyUI and, for speed, an NVIDIA GPU; everything also runs on the CPU, slowly.

cd ComfyUI/custom_nodes
git clone https://github.com/PedroMarinhoDev/ComfyUI-SceneCut
pip install -r ComfyUI-SceneCut/requirements.txt

example_workflows/SceneCut-basic.json

output/scenecut/<video>/scenes_00001/
  <video>_001.mp4  <video>_002.mp4  …
  scenes.json

The advanced workflow splits Detect into Detector Loader → Detect (Advanced) → Refine → Validator Loader → Validate (Advanced), each with every parameter.

Export

Exports run in the background with a progress bar; you can keep reviewing meanwhile. Frames are found by their timestamps, so a clip in the middle of a two-hour film is one keyframe seek away, and every clip is checked to hold exactly its scene's frames.

Performance

Measured on 2026-10-16 with an RTX 4090 and an Intel Core i9-14900K. "End to end" is the basic workflow: detection, validation, the viewer's preview copy and thumbnails. Export writes every scene with its sound, H.264 at high quality on the CPU, and every clip was checked to hold exactly its scene's frames.

FAQ