Online Tool Store
· 4 min read
How to Find Every Cut Point in a Video
Heshan Fernando
Co-founder & COO
You’ve been handed a forty-minute recording and asked for chapter markers, or a set of clips, or just a list of where the shots change. Scrubbing through a timeline looking for cuts is exactly the kind of work that feels productive and isn’t.
A scene change detector does the first pass in seconds. What it can’t do is tell you which of its detections are meaningful — that part still needs you, and knowing why it makes mistakes is what makes the output usable.
How detection actually works
The basic method is simple: compare each frame with the one before it, measure how different they are, and flag the moments where the difference spikes.
A hard cut produces a large spike, because two consecutive frames from different shots share almost nothing. That case is easy and detection is reliable.
Everything else is harder. A cross-fade produces a run of moderate differences rather than one spike, so catching it means lowering the threshold — which then also catches fast camera pans, lighting changes, and anyone walking close past the lens. A camera flash produces an enormous single-frame difference that isn’t a cut at all.
That’s why a minimum scene length setting exists. It collapses clusters of detections into one, which fixes flashes and multi-frame transitions in a single stroke.
Why people get stuck here
- Sensitivity as a single dial. Turn it up to catch fades, get false positives on motion; turn it down, miss the fades.
- Flash frames. A single bright frame reads as two cuts, one in and one out.
- Long files. Reading every frame of a long video is slow, whatever the tool.
- Timecode format mismatches. Chapters, editors and clip exporters each want a slightly different format.
What good detection output looks like
A minimum scene length
Nothing under a second or two is usually a real scene. Setting a floor removes the majority of false positives without touching genuine cuts.
Grouped transitions
Two detections a frame apart are one transition, not two. Good output merges them and says so.
Timecodes you can use directly
A list you can paste into a chapter field, or a CSV of start and end points for a batch export. Reading timecodes off a screen and retyping them is where the time savings evaporate.
| Setting | Low Sensitivity | High Sensitivity |
|---|---|---|
| Catches | Hard cuts only | Cuts, fades, dissolves |
| False positives | Few | Motion and lighting changes |
| Best for | Interviews, static shots | Edited montage, mixed footage |
Common mistakes to avoid
- Running the detector at maximum sensitivity on footage with a lot of camera movement, then working through hundreds of false detections by hand.
- Setting the minimum scene length longer than your actual shortest shot, which silently merges real cuts.
- Trusting detected boundaries as exact edit points — a cut detected on frame 4,102 may actually belong on 4,101.
- Loading a two-hour file into a browser-based tool and expecting instant results.
- Using scene detection on screen recordings, where slide transitions and scrolling behave nothing like camera cuts.
How to do it with Scene Change Detector
The Scene Change Detector reads frames locally, so the video never leaves your device.
- Add the video file. Length is limited by your device’s memory rather than an upload cap.
- Start with low sensitivity to find the hard cuts, and raise it only if you’re missing fades.
- Set a minimum scene length that matches your shortest genuine shot.
- Export the timecode list or CSV, and spot-check three or four detections against the footage.
Other video utilities that run in the browser are in the tools directory.
Frequently asked questions
Why does a camera flash register as a cut?
Because the frames genuinely differ enormously for one frame. The detector is measuring exactly what it’s designed to measure. A minimum scene length collapses those single-frame spikes, which is what that setting is for.
Will it find cross-fades and wipes?
Partly. Gradual transitions produce a run of moderate differences rather than a spike, so they need higher sensitivity — at the cost of more false positives on motion. Shot transition detection is an active research area precisely because the gradual cases are hard.
Is the video uploaded?
No. Frames are decoded in the browser, which is also why long files are slow.
Final thought
Run it at low sensitivity first. A clean list of hard cuts that you extend by hand beats a hundred detections you have to filter, every time.