Private video tools

Chapters and scene cuts, measured on your device

Find where a clip changes shot, check every boundary against its own thumbnail, then export a chapter list for a video description or a WebVTT chapter track for a player.

This one is not AI. It compares colour histograms between consecutive sampled frames and calls a large enough change a cut. Nothing was trained, nothing is fetched, and it runs in seconds on any device. It also cannot tell you what a chapter is about, which is why you write the titles.

Result proof

A chapter list whose every marker you have seen

01

Input

Any video file your browser can decode

02

Local pass

Colour-histogram comparison between sampled frames, with a minimum chapter length

03

Result

Chapter list and WebVTT chapter track

How to verify it: Compare each thumbnail against the timestamp beside it, scrub the source player to a marker and confirm the shot changes there, then paste the list into a description or load the .vtt into a player and open its chapter menu.

What people use this for

Chapters for a long upload

Get the seams of a tutorial, lecture or stream marked in seconds, then name the sections yourself rather than scrubbing for them.

A shot list for an edit

Turn a rough cut into a numbered list of shot boundaries you can work against, without a timeline export.

Checking a recut

Run it on two versions of the same piece and compare where the cuts landed.

Feeding a highlight pass

The same boundaries drive Auto Highlights, so the scenes you review here are the ones it scores.

Limits worth knowing

  • It measures colour change, not meaning. A cut between two similar-looking shots is invisible to it, and a lighting change is indistinguishable from an edit.
  • Cross-fades and dissolves have no single frame where the colours jump, so they are usually found late or not at all.
  • Sampling is roughly one frame per second and capped at 120 samples, so a boundary is accurate to about a second and a long clip is sampled more coarsely.
  • A minimum chapter length is always applied, so two cuts inside the same second collapse into one.
  • It cannot name a chapter, because nothing in it reads the picture, the audio or a transcript.

How the local workflow runs

  1. Choose a video. It stays on your device.
  2. Pick a sensitivity: fewer and longer chapters for a single-location shoot, more and shorter for a fast-cut edit.
  3. Run the detection and read the thumbnail strip.
  4. Remove any boundary that turned out to be a lighting change, and type a real title over each number.
  5. Copy the chapter list, or download it as a .txt or a .vtt chapter track.

Frequently asked questions

Is this AI?

No, and it is worth being exact about it. The tool samples about one frame per second, builds a colour histogram for each sampled frame, and measures the chi-squared distance between neighbours. A distance above the sensitivity threshold is a cut. Nothing was trained, nothing is downloaded, and the tool has no idea what is in your video. An earlier version of this shipped under an AI label and it was not AI then either.

Does the video leave my device?

No. The file is read by your browser, decoded by your browser, and measured by your browser. There is no upload step and no server involved.

Will it name the chapters for me?

No, and it will not pretend to. Nothing here reads the picture, the audio or a transcript, so it has nothing to name a chapter with. The chapters arrive numbered and you type what each section is. The timestamps are measured; the words are yours.

Why did it miss a cut?

Almost always because the two shots look alike. Two camera angles on the same person in the same room share a colour distribution, so there is no change for the detector to see. Cross-fades and dissolves are the other case: they change gradually over many frames, so no single pair of frames jumps far enough to register.

Why did it mark a cut where there is no edit?

A lighting change, a cut to black, a camera flash, or a pan onto a bright window all change the colour distribution wholesale. The measurement is correct and the interpretation is wrong, which is why the tool shows you a thumbnail for every boundary instead of just handing over a list.

How accurate are the timestamps?

The clip is sampled at roughly one frame per second, capped at 120 samples, so a boundary is located to within about a second and a long clip is sampled more coarsely than a short one. Seconds are floored rather than rounded, so a marker always sits on or just before its own cut rather than inside the shot that follows.

Why does it say my chapter list will not work on YouTube?

YouTube requires the first timestamp to be exactly 0:00, at least three chapters, and every chapter to run at least ten seconds. It enforces all three by doing nothing: a description that fails any of them renders as plain text with no chapter menu and no explanation. The tool checks them and names the one that failed, so you find out here rather than after publishing.

What is the WebVTT file for?

It is a chapter track, one cue per chapter, in the format video players read to build a chapter menu. It has none of the YouTube description rules, so it works for clips that cannot meet them.

Ready to try it?

No upload, no model download, no invented chapter titles.

Find the scene cuts