How to turn a YouTube video into Shorts, Reels and TikToks (without re-editing it)

You have a 40-minute upload. Somewhere in it are five or six moments that would do well as Shorts. The old way of finding them is to scrub the timeline with one hand and a notepad in the other, then cut each one by hand, then reframe each one to vertical, then caption each one.

The whole point of an AI clipping tool is that you paste the link and skip all of that. This post is about what happens in between, because the details decide whether the clips are usable.

The two ways in: a link or a file

Paste the YouTube URL. Fastest, and the right choice for anything already published. The tool fetches the video itself, so a 2GB source never has to travel up from your laptop.

Upload the file. The right choice for anything not on YouTube yet, or if you have the original export and want native quality. An upload is the file exactly as you rendered it, so if you rendered 4K, the clips have 4K to work from.

One thing to know about links, because it isn't obvious and most tools don't tell you: the resolution a tool can pull from YouTube is not automatically the resolution you uploaded. Cloud services fetch from datacenter IP addresses, and YouTube serves those a 360p stream and refuses the HD ones. A clip made from a 360p source and stretched to 1080x1920 vertical is a 5x upscale. It looks exactly like it sounds.

We route the HD fetch differently so link imports come through at 1080p or 1440p, and only the seconds we actually clip are fetched in HD, which is what keeps it affordable. If a tool's link imports look soft and you can't work out why, this is usually why.

What the tool does with it

Roughly four stages, and each one has a way to go wrong that you can spot in the output.

1. Transcription. Every word, with a timestamp. This is what everything else reads, so its accuracy is the ceiling for the whole run. Names get mangled; if a clip's caption spells your guest's name three different ways, the transcript did.

2. Finding the moments. A language model reads the transcript and picks the segments with a hook, a build and a payoff. The better tools classify the video first: a podcast clip is a complete thought, a comedy clip is setup plus punchline plus the reaction after it, an educational clip is one insight. Get that wrong and you get punchlines with no setup. We measured our own earlier version doing exactly that: 7 of 10 clips ended mid-sentence.

3. Reframing to 9:16. The 16:9 frame has to become vertical, which means throwing away 56% of the width. Where the crop sits is the whole question. A centred crop of a two-person shot lands on the gap between them. Read how auto-reframe works if your clips keep showing a shoulder and an empty chair.

4. Captions and render. Word-timed captions burned in, one file per clip, ready to post.

Doing it on GetClipMachine

1. Open the Studio and paste the YouTube link (or drop the file).
2. Wait. A 10-minute video takes a few minutes; a 60-minute one takes about the same for the finding step because we process long videos in 15-minute windows, plus render time for however many clips it picked.
3. You get up to 12 clips, each with a title, a one-sentence reason it was picked, and a score. Read the reason before the score; a specific reason ("the guest names a figure the host didn't expect") means the model found something. A generic one means it didn't.
4. Download the ones you want. Free accounts get 3 videos a month at 1080p with a watermark; Pro is €15/month, no watermark, 4K export, videos up to 180 minutes and 4 GB.

Clips come out between 20 and 45 seconds by default, never over 60. If you're wondering why that range, we wrote up the reasoning.

What to check before you post

• Does the first line stand alone? Play the first two seconds with the sound off and read the caption. If it's the middle of a sentence, the clip started late. Every tool gets this wrong sometimes; the good ones get it wrong rarely.
• Is the speaker in frame for the whole clip? Especially on cuts between camera angles. A crop that was right for the wide shot is wrong for the close-up.
• Did the reaction survive? On anything funny, the laugh after the line is the payoff. A clip that ends on the punchline itself is a joke told to silence.
• Is the caption spelling right? Names and product names. A transcript can be 98% accurate and still get the one word that matters wrong.

When not to bother

A few kinds of video don't clip well from a link, whatever tool you use:

• Screen recordings and slides. There's no face to frame and the text is unreadable at vertical crop widths. Clip these by hand or don't.
• Music and montages. The transcript is empty, so the moment-finding step has nothing to read. We flag audio-only uploads outright rather than producing twelve clips of nothing.
• Anything under about two minutes. It is the clip. Reframe and caption it yourself.

For everything else, and especially for the long-form talking-head video that most YouTube channels are built on, the link is enough. The YouTubers page covers how to use your retention graph to sanity-check what the tool picked.