Format Converter
Free toolNo account

Video Scene Detector

Find likely hard cuts in a selected video range. Adjust the detection threshold, jump to candidates in the source preview, and export a timestamp list as CSV or JSON.

Detected cut candidates will appear here.

Spend less time scrubbing for edit points

Scan locally without uploading the clip
Locate hard cuts in a finished montage when the edit project is unavailable. Use the list as an index, then preview each position before treating it as a true boundary.
Find promising places to sample reference images. A cut timestamp can guide selection, but choose a nearby settled frame instead if the first image after the change is blurred.

video scene detector

Use a video scene detector to build a reviewable cut list

A video scene detector can turn a long visual search into a short review list. This page compares nearby decoded frames and reports substantial changes as candidates. It does not decide where the story changes or split the file automatically; you can inspect the reported position and make that decision in context.

Search focus
video scene detector
Input
Video file
Output
CSV / JSON cut list
Processing
Browser local

Scan, review, and export candidate cuts

Choose the actual clip whose time axis matters. A rendered export, a trimmed copy and an editor timeline may all have different starting points, so keep track of which file supplies the reported timestamps.

  1. 1

    Enter the source range

    Set a start and a length in seconds. A start of 35 with a length of 20 requests the portion from 35 to 55 seconds. If the file ends earlier, the scan stops there. The first decoded frame is a baseline rather than a reported cut.

  2. 2

    Set a threshold and scan

    The threshold sets the minimum scene-change score for a candidate. A lower setting accepts smaller visual changes, with more false positives possible. A higher setting keeps stronger changes but can miss edits between two similar compositions. Changes to the settings affect the next run.

  3. 3

    Verify and share the cut list

    Select a timestamp to open the original video preview at that position, if the browser can play it. Compare the image just before and after the candidate. Download CSV for a spreadsheet or JSON for the scan settings, candidates and truncation flag. An empty report means the scanned frames produced no qualifying candidates.

A picture change is not always an edit

A hard cut often produces a large change between adjacent pictures, but a full-frame flash or a sudden pan can do the same without an edit. Conversely, two shots with similar framing and lighting can be separated by a real cut while looking nearly identical to a simple detector.

This page uses FFmpeg's scene score on a reduced-size image, not object recognition or a semantic video model. The score is a comparison measure on a scale from zero to one, not a percentage probability. Review the visual evidence before using a reported time in a publishing or editing workflow.

Threshold, scan length and timestamp behavior

The threshold ranges from 0.05 to 0.90. Each window can be 0.1 to 120 seconds, and the tool scans the first regular video track. Reported positions are rounded to milliseconds on the source timeline, not converted to an editor's frame or timecode format. The colored marks show candidate locations across the requested window; if the source ends early, the unused span is not a proof that the video continued.

When a candidate list is useful

Locate hard cuts in a finished montage when the edit project is unavailable. Use the list as an index, then preview each position before treating it as a true boundary.
Find promising places to sample reference images. A cut timestamp can guide selection, but choose a nearby settled frame instead if the first image after the change is blurred.
Share review notes with an editor or annotation team. Export source-relative seconds and scores rather than copying positions from a player by hand.

Review around the timestamp, not just at it

The candidate is associated with a changed decoded picture. Browser seeking may land slightly differently depending on the format, timestamp precision and player. Scrub a little before and after the marker, and verify the exact frame in an editor if a split must land on a frame boundary. Do not derive a cut from the score alone.

  • Start around 0.30. If fast camera motion, exposure changes or a flash produce extra candidates, raise the threshold and rerun the same interval before choosing which to keep.
  • If a visible edit is missing, lower the threshold and shorten the window around it. Much of the surrounding stretch may be irrelevant, and a shorter scan makes false positives easier to compare.
  • Overlap consecutive scan windows slightly when reviewing a longer video. The first decoded frame establishes a baseline; placing a candidate exactly at the start can leave it without a preceding picture for comparison.

Local scan and a separate report

The browser downloads the processing engine when necessary, then decodes the selected clip in a local worker. The video is not uploaded or rewritten by this tool. Cancel or reset can stop a scan without changing the source file.

The report includes timestamps and scores, not images, audio or the original filename. JSON also records the requested settings. Timestamps can still reveal something about the edit structure, so share them only with the people who need to review the work.

Common sources and their tradeoffs

For a six-minute presentation with slides and a camera feed, scan short windows around expected switches. High-contrast slide changes may register even when the editor did not cut to another shot.

For a fast product montage, run at the default threshold and inspect candidates in source order. Raise the setting if whip pans dominate the results, then compare what disappears.

For a clip beginning with a fade, include a few seconds beyond the introduction. The detector may ignore the gradual fade while still reporting an abrupt cut to the first full shot.

Boundaries and preview compatibility

This is a single-file tool for local clips. It returns up to the first 300 cut candidates per scan and flags a report that hits additional candidates. Use a shorter window or raise the threshold when the list is cut short. Detection depends on the browser engine being able to decode the source. Native video playback is separate; if the preview is unavailable, use an editor to inspect the CSV timestamps.

  • Use a shot-by-shot edit review for dissolves, fades, subtle match cuts or footage with aggressive strobing. These changes do not reliably correspond to a single strong frame difference.
  • Do not use a cut list as a scene transcript or a ready-to-import edit decision list. The report has no dialogue, clip names, timeline frame rate or drop-frame timecode.

Questions about video cut detection

Can it split the video for me?

No. The result is a timestamp report. After verifying the boundaries, enter the desired range in a video trimming tool or use an editor if you need many separate clips.

Why did it miss a dissolve or fade?

A gradual transition spreads the change over several frames. None may score high enough individually. Lowering the threshold can help with some examples but also creates more false positives, so review gradual transitions manually.

What does a report with no cuts mean?

The scan decoded a valid range but did not see a frame difference at or above the selected threshold. A steady shot may legitimately return nothing; this does not rule out subtle edits or changes outside the chosen window.

Why does a timestamp differ from my editing timeline?

The position is measured in seconds relative to the file you supplied. A trimmed export, a timeline with an offset or a different timecode scheme can use different numbers. Reference the same source and compare the neighboring frames.