Scene Detector: Split a Video into Shots (Timecodes, EDL & Clips)

About the Scene Detector: Split a Video into Shots (Timecodes, EDL & Clips)

Open a video and the page finds every shot change: each hard cut, where one picture replaces another from one frame to the next, and each fade to black between two scenes. You get a shot list with SMPTE timecodes at the video’s own frame rate, a thumbnail of every shot, and a player to check each cut, add one that was missed, move one by a single frame or remove a false one.

With a Pro pass, the shots become files for your edit: a CSV shot list, a CMX 3600 EDL that editing apps import as a timeline, YouTube chapter timestamps, a contact sheet as PNG or PDF, and every shot as its own clip in a ZIP, copied without re-encoding or cut on the exact frame. The free preview finds every cut and shows the shot count with the first shots of the list. The video never leaves your device: it is read, frame by frame, in your browser.

How the cuts are found

Every frame is drawn 96 pixels wide and turned into hue, saturation and brightness. The change from one frame to the next is the average difference of the three; a frame is a cut when its change is several times the change of the frames around it. Comparing with the frames around it is what keeps a fast pan or a handheld shot, where every frame changes a lot, from being cut to pieces. The sensitivity sets how many times larger the change must be, and how large at least.

A fade to black is a run of frames darker than a threshold between brighter ones; the new shot starts in its middle. Two cuts closer together than the shortest shot keep the stronger one, so a camera flash or a single bright frame does not become a shot of its own. A cut shows in the list as the first frame of the new shot.

Timecodes and the EDL

Timecodes are written as hours:minutes:seconds:frames at the video’s frame rate. For 29.97 and 59.94 fps you can choose drop-frame timecode, which skips frame numbers 00 and 01 (00–03 at 59.94) at the start of every minute except every tenth, so the timecode keeps pace with the clock; drop-frame is written with a semicolon before the frames on the page and in the CSV.

The EDL follows the CMX 3600 format: a TITLE: and an FCM: line, then one cut event per shot with its reel, tracks, source in and out, and record in and out, where out points are the first frame after the shot. The file starts at the timecode you set (00:00:00:00 for a file without its own timecode) and the timeline at 01:00:00:00 unless you change it. With Name the clip, each event carries a “* FROM CLIP NAME:” note with the file’s name, which editors use to link the events to your video; shot names go in as “* SHOT NAME:” notes. Strict CMX readers take up to 999 events; the page says when a list is longer.

Clips: lossless or exact

Lossless clips copy the video data as it is, so they keep the original quality and are made in seconds. A copy can only start and stop on a keyframe (the frames a video can start from), so a lossless clip runs from the last keyframe at or before its cut to the first one at or after the end of its shot: every frame of the shot is in it, and it may begin and end with a few frames of the shots around it. Shots with no keyframe between their starts share one clip, named like video-shots-007-009.mp4. Many encoders put a keyframe at each cut anyway.

Exact clips start and end on the frames of their shots: the picture is encoded again as H.264 MP4 at the quality you choose, which takes much longer. Sound is copied where the clip’s container holds it.

YouTube chapters

The timestamps are one line per shot, “0:00 Shot 1”, or the names you gave the shots. YouTube shows chapters when the first timestamp is 0:00, there are at least three in ascending order and each chapter is at least 10 seconds long (YouTube Help: video chapters). Join shots shorter than 10 s joins short shots to the chapter before them, so the list is one YouTube accepts; for a talk or a tutorial, a low sensitivity gives one chapter per slide or scene.

Limitations

• Reading the video needs WebCodecs to decode its frames: a current Chrome, Edge, Safari or desktop Firefox. HEVC (H.265) video needs a hardware decoder, which most recent computers and phones have.
• Videos of up to 2 hours. Every frame is decoded, so a long film takes a while; keep the page open until it finishes.
• Slow dissolves, fades to a colour other than black and flashing lights can be missed or found twice. Check the list in the player and fix those cuts by hand.
• The page cannot play every file in its own player (for example MKV in Safari); then it shows the frames one at a time with a slider instead.
• Lossless clips start and end on keyframes, so they can begin a few frames early and end a few frames late; choose Exact for clips cut on the frame.

Privacy

On this device: nothing is sent

Your video is opened and read in your browser and is never uploaded. The shot list, the names you give shots and the exports are made on your device; only your settings are remembered in this browser.

Open-source components

This page can load these open-source components, each only when a task needs it:

• It uses FFmpeg under the LGPLv2.1, through the Mediabunny add-ons: licence and source code.
• It uses the LAME MP3 encoder under the LGPL, through the Mediabunny MP3 add-on: licence and source code.
• It uses Mediabunny under the MPL 2.0: licence and source code.
• It uses the TurboRes ProRes decoder under the MPL 2.0: licence and source code.

You may replace them with your own builds. Their licences and the credits of every library the site uses are on the open-source credits page.