From a Long Clip to a Highlight Reel — Automatically
How automatic highlight detection scores scenes, picks the best moments to a target length, and stitches them into a real, playable clip — entirely in the browser.
You have a ten-minute clip and need a thirty-second cut for social. Doing that by hand means scrubbing the timeline, marking in/out points, and stitching segments. Automatic highlights does the first pass for you: find the moments worth keeping and assemble them into a real, playable reel.
Step 1: Find the scenes
Before you can pick highlights, you have to break the video into candidate moments. Sample one frame per second and compute a colour histogram for each. When two consecutive histograms differ sharply (a chi-squared distance spike), that is a scene cut. Group the frames between cuts and you have a list of scenes with start/end times and a representative thumbnail each.
Step 2: Score the moments
Not all scenes are equally interesting. Each gets a score from cheap, meaningful signals:
- Visual detail — high-variance frames (action, faces, texture) score higher than static ones.
- Pacing — shorter scenes often indicate cuts and energy, so they get weighted up depending on the chosen focus.
You can bias the scoring toward speech, motion, subject, or a mix, depending on what kind of reel you want.
Step 3: Pick to a target length
Now it is a budget problem. Sort scenes by score, then greedily add them until you hit your target duration (15, 30, or 60 seconds), allowing a slight overshoot so you do not cut a moment off mid-beat. Finally, re-sort the chosen scenes back into chronological order so the reel reads naturally.
Step 4: Actually encode the reel
This is the step that makes it a tool rather than a suggestion. With the chosen time ranges in hand, trim each range from the source and concatenate them into a single clip using an in-browser FFmpeg build. The output is a real MP4 you can play and download — not just a list of timestamps you would have to cut yourself.
Getting better results
- Pick the right target length. A 15-second cut is punchy; 60 seconds tells more of a story. Match it to the platform.
- Review the cuts. Automatic selection is a strong first draft, not a final edit. Trim a beat here, drop a scene there.
- Add captions or music after. The reel is your skeleton; titles, captions, and a track give it life.
Because detection, scoring, trimming, and encoding all run locally, even a long source clip stays private — nothing is uploaded to find your highlights.