YouTube Shorts Maker

What this does, and what it doesn't

There are two jobs inside "make a Short from my video". One is mechanical: trim, crop, caption, export. Your phone does that, and so does every editor you already own. The other is judgement: out of fifty minutes, which forty seconds will hold a stranger who has no idea what came before?

That second job is the one that takes the time, and it is a text problem rather than a video problem — which is why it can be done here at all. This tool reads the captions, finds passages that stand alone, checks the timecodes against the actual video, and hands you a cut list.

We deliberately don't do the video half. Editing in a browser means a large download, slow encoding and frequent failures on phones — and you'd get a worse result than the editor already on your device.

Doing it by hand, and why nobody keeps doing it

The manual version is not complicated, it is just long. You scrub the video looking for a bit that might work, note the time, lose your place, scrub back, decide it needed the sentence before it, and start again. An hour of source takes the better part of an hour to mine, and the reward is four clips.

That is why channels sitting on years of good long-form post so few Shorts. The material is there; the search is what nobody has an hour for, every week, forever. Reading a transcript takes a machine seconds, and the judgement about what stands alone turns out to be mostly a judgement about text.

How to use it

1. Paste a long video. Your own podcast, lecture, interview or tutorial — anything with captions and more than a couple of minutes in it.
2. Wait for the read. It fetches the transcript and works through it. A long video takes a minute or so, and it tells you which stage it's on.
3. Open each moment. Every clip links to that exact second so you can hear it before you commit to cutting it.
4. Take the cut list. Copy it, or download it as text, as a spreadsheet, or as an ffmpeg script that cuts the clips from your own copy of the video.

What makes a passage work on its own

A Short has no runway. Someone arrives mid-scroll with no idea who you are, and the first second either earns the next one or doesn't. That rules out most of a long video, and the parts it rules out are predictable:

• Anything that leans backwards. "Like I said earlier", "going back to that point" — the viewer wasn't there for the earlier bit.
• Introductions and sign-offs. They're about the video, not about anything.
• Sponsor reads and housekeeping. Obvious, and still the most common thing people clip by accident when they cut on timecode alone.
• Passages that are only loud. Energy isn't a point.

What survives is usually a complete thought: a claim with its evidence, a question with its answer, a short story with a turn, a specific number, or a clearly stated disagreement. Those are the shapes the tool looks for.

How long a clip should be

Shorts can run to three minutes now, which is a trap. The clips here sit between 20 and 90 seconds, and the target is 30 to 60 — not because there is anything magic about a minute, but because that is how long a complete thought usually takes to say.

Under twenty seconds you rarely get a whole one: a claim with no support, or a punchline with no setup. Past ninety you are not making a Short, you are making a short video, and the thing that made the passage work — that it resolves before anyone's attention runs out — has gone. The length is decided by where the thought ends, and then checked against those bounds rather than the other way round.

The timecodes are checked, not trusted

Language models are good at spotting a self-contained passage and careless with arithmetic. So every timecode that comes back is verified against the video before you see it: a clip that runs backwards, starts past the end, runs shorter than 20 seconds or overlaps one already chosen is dropped, and anything over 90 seconds is trimmed to fit.

That's also why you get four to eight clips rather than twenty. A video that genuinely contains twenty standalone moments is rare, and a list padded to look generous would just cost you the time of checking them.

Check before you cut

Every clip links to its exact second, and that link is the most useful thing on the page. Open it and listen to the first three seconds. If the opening sentence needs something you did not just hear, skip the clip — you have spent ten seconds instead of finding out after editing it.

Two things worth listening for that text cannot tell you. Whether the speaker is actually on screen at that moment, because a vertical crop of a slide is not a Short. And whether the sentence starts cleanly, since captions do not record a false start, a cough or someone talking over the top.

Getting the actual files

The cut list downloads three ways. Plain .txt to keep beside the project, .csv to open in a spreadsheet or import as markers, and an ffmpeg script that does the cutting for you.

That last one is the shortest path from this page to finished vertical files. Put your own copy of the video in a folder, drop the script beside it, run it, and you get one MP4 per clip, already cropped to 9:16 and named after its title. It starts from your file — nothing in it fetches anything from YouTube, which is both the honest arrangement and the one that actually works.

If the speaker isn't centred in frame, the crop filter in the script takes an x offset — the header explains where. And if you'd rather cut by hand, every clip carries a link that opens the video at that exact second.

You don't have to take the whole list either. Each clip has its own Copy the cut command button, which gives you the single ffmpeg line for that one clip — the same line the script would have run. Paste it into a terminal in the folder holding your video and you get that clip and nothing else, which is what you want when only one of the eight is worth making.

Then what?

Cut the clip at the times given, crop to 9:16 keeping the speaker in frame, and add captions — most Shorts are watched without sound. Then the metadata: the hook belongs at the top of the description, and the description generator puts the rest in the order YouTube reads it. Check the title against length and keyword position with the title generator, and if you want to see what your niche is tagging, the hashtag generator counts what recurs across videos you choose.

Once it's published, the Shorts thumbnail downloader gets you the full vertical frame so you can see what it looks like everywhere the Short is listed rather than played.

Videos this won't help with

It needs captions, so a video nobody has captioned has nothing to read — the transcript generator will tell you quickly whether one exists. It also needs length: a video already close to Short length has nothing to cut down. And a video whose whole value is visual — a silent build, a drawing, a gameplay run — leaves no trace in the captions, so the text can't tell you where the good parts are.

Frequently asked questions