How the tool finds cuts
The tool decodes small 64×36 frames, up to 8 each second. For each frame, it measures two changes from the frame before it:
- The change in brightness of each pixel. This finds cuts between two shots with similar colours.
- The change in the colour histogram. This stays low when the camera moves, so a pan is not a cut.
The tool mixes the two values into a score from 0 to 1. A frame with a score above the threshold is a cut. The chart shows each score, so you can see why the tool marked a cut. For the theory, see shot detection.
Choose the settings
| Setting | What it does | When to change it |
|---|---|---|
| Sensitivity | Sets the threshold. Higher sensitivity finds more cuts. | Move it up for fades and soft cuts. Move it down if fast motion causes false cuts. |
| Minimum scene length | Ignores a cut that comes too soon after the last cut. | Make it longer for flashes, strobes, and fast edits. |
The tool keeps the frame scores in memory, so a change to a setting updates the scene list at once.
Export formats
- JSON: the scenes with start, end, and duration, for scripts and apps.
- CSV: one row for each scene, for spreadsheets.
- EDL (CMX 3600): a timeline for Premiere Pro, DaVinci Resolve, and Final Cut Pro.
- YouTube chapters: a list of times to paste into a video description. YouTube needs at least three chapters, the first at 0:00, and each one at least 10 seconds long. The tool joins short scenes to meet these rules.
- FFmpeg script: one command for each scene, which cuts the video into one file per scene.
What you can do with scenes
- Split a long video into clips for social media.
- Make a storyboard with our frame extractor, or show the scenes to an AI with Video to AI.
- Measure the pace of an edit: the number of cuts per minute and the average shot length.
- Index a video library by scene. The video understanding API adds speech, text, and objects to each scene.