Video tool

Video Auto Clip

Point it at a video and it finds the moments worth keeping, scores them against a hundred points, and cuts them into separate MP4 files. Each clip downloads on its own as soon as it is ready, so you do not wait for the whole batch. The video is read, measured and rewritten inside this page and is never sent anywhere.

1. Choose a video

Your video is opened by this page alone. It is never uploaded, and closing the tab forgets it.

Choose a video, or drop one here You can also paste a video with Ctrl or Command and V MP4 or WebM, up to 2 minutes

Choose a video to begin. Nothing is uploaded.

2. What it found

The chosen moments appear here, drawn over the places where somebody is speaking.

3. Check a clip before you take it

The best clip starts playing by itself once the scan finishes, muted. Turn the sound on with the button below it. It plays the original video at the clip boundaries, so it is instant. Creating the clip is the slow half and only happens when you ask for it.

How the choice is made. The sound is measured in short windows, which is what tells the tool where somebody is speaking and where they pause. Moments are built around those pauses, scored out of a hundred, and the cut is then moved to the quietest point near the edge and given a small safety pad. The picture is measured too, and scene changes are marked on the strip in red, but they do not decide anything on their own. That is the difference from a scene cutter: a fast-cut gaming video gives a few sensible clips here rather than dozens of two-second fragments, and a single-camera talker with no cuts at all still gives you real highlights.

What it measures, and what it does not

The score on each card breaks down into the parts that were actually measured. Open Why this clip to see them.

Complete structure

Starts and ends in silence, is not cut in half by a shot change, and sits near the length you asked for. This is a structural reading, not a judgement about meaning.

Speech quality

How much of the moment is speech rather than dead air, and how steady the delivery is.

Clean boundaries

Whether the edges land in silence. This is the part that stops a word being cut, and it is why the moment is placed on a pause rather than on a scene change.

Visual activity

How much the picture moves. A still frame and a shaky camera both score badly here.

Audio dynamics

How much the loudness varies inside the speech, measured only over the spoken parts.

Distinct from the rest

How little it shares with the clips ranked above it. Near-duplicates are suppressed and the stronger one is kept.

What this tool does not do

The honest limits, with the trade-offs in the open rather than in a footnote.

It does not understand the words. Moments are found from the shape of the sound and the picture, never from what is being said. Two moments that mean very different things can score identically if they are said and shot the same way.
It does not reframe or resize. Every clip is cut from the original picture, at its original size. If you need a vertical clip for a phone or a square one for a feed, crop it afterwards, in whatever tool you already use for that.
It does not follow a subject around. A centred subject stays centred, because the crop would be centred if there were one. Tracking a face across the frame is a different job.
Music with no speech finds nothing. It will tell you that rather than cutting the file at arbitrary points. A video with speech but no strong single moment falls back to the clearest shots, and says that is what it did.
Nothing is uploaded and nothing is asked of a server. The whole thing runs in this page. There is no account, so there is nothing to send your video to, and no model is downloaded because none is used.
Long videos take a while. Every clip is encoded on this machine, which is why clips appear one at a time rather than all at the end, and why a two-hour file is a poor choice for this page.

When there is nothing to find

Music with no speech, or a silent video, produces no highlights, and the tool says so rather than cutting the file at arbitrary points. A video with speech but no strong single moment falls back to the clearest shots, and says that is what it did.

When to use something else

To keep particular pages, split the video yourself first: this tool chooses moments, it does not obey a page list. If the goal is a smaller file, cutting does not help. If a clip has to be signed, add the signature afterwards so it covers the finished file.

How automatic clipping decides where to cut

The tool looks for changes in the video rather than understanding what is in it. Strong cut points are usually where a great deal changes at once: a new scene, a big movement, or a clear change in brightness. Long takes with a steady camera produce few candidates, because there is little to detect.

That means the output is a set of suggestions rather than an edit. Two clips that look similar to you may be merged because the tool cannot tell them apart, and a single scene with fast movement may be split more than once. Review the clips before using them, and expect to adjust.

Longer minimum clip lengths produce fewer, more usable fragments. Very short fragments are rarely useful unless you are looking for individual beats, so start with a longer minimum and reduce it only if you are losing the moments you wanted.

Everything runs in the tab, so the length of the video is limited by your machine rather than by an upload quota. An hour-long recording will take noticeably longer than a short clip, and a phone with limited memory will struggle with a very large file regardless of how fast its processor is.

Further reading: Read the full guide: Video Auto Clip

Common questions

Is my video uploaded?

No. The video is read in this tab, analysed on your machine, and the clips are written back to your device. Nothing is uploaded, and no account is needed because there is nowhere for the file to go.

How does it decide what to clip?

It scores the video out of a hundred on speech quality, how complete a thought is, how clean the boundaries are, visual activity and audio variation, then picks the strongest stretches that are not too close together. Each clip is then moved to the quietest moment near its edge, which is what stops a word being sliced in half.

How long is each clip?

You choose: 15 to 30 seconds for a quick answer, 30 to 60 for a point made once, or 60 to 90 for a longer explanation. The tool cuts to your setting, it does not impose one.

What if it finds nothing worth clipping?

It will tell you rather than produce a clip anyway. That normally means the video has little speech, or is mostly music or silence, and there is no highlight to find. It is a sign to use the video as it is.

Does it add captions or text?

No. It finds and cuts moments and leaves the audio alone apart from optionally dropping it. Nothing is written over the picture, so what you get is the video you had, shortened.