How I Turn a YouTube Video Into Shorts Without Editing

To turn a YouTube video into Shorts without editing, I paste the link into Claude Code and say “clip this video.” It downloads the captions, finds the pauses, picks three moments worth posting, cuts each one into a vertical clip with the disclosure burned on, and writes the captions. I watch each clip and approve or reject it. This post walks through how it works and gives you a generator that writes the same system for your videos.

This post contains affiliate links. If you sign up through one, I earn a commission at no extra cost to you. I only point to tools I actually use or would recommend.

The videos were already done

I film product reviews for the Amazon Influencer Program. Short ones, usually around two minutes, me holding the thing and saying what I think of it. I’ve filmed well over a thousand of them, and until last week not one had ever been cut into a Short or a TikTok.

Editing is just not really my thing. Every time I thought about short videos, it felt like pressure to create new videos, on top of the ones I was already making. So I didn’t. I was frozen on it for years, and the videos kept piling up in one place.

What changed the math is that these were already done. All that was missing was someone to find the twenty good seconds in each one and crop them to fit a phone, and I wouldn’t have found the time to do that myself. So I had Claude Code do it, and it’s turned into an easy way to get my videos more attention without filming anything new.

What does it take to turn a YouTube video into Shorts?

Less cutting than you’d think and more picking. The cutting is a handful of commands. The work is deciding which twenty seconds out of two minutes are worth posting, and that needs the transcript and the frames together, because a great line over a boring shot isn’t a clip.

One video goes through this, in order:

1. Download the video and its captions. 2. Turn the captions into a transcript with timestamps, and find every gap in my speech. 3. Pull a frame every few seconds into one contact sheet, so the shot changes are visible. 4. Screen the transcript and the frames for words platforms bury videos over. 5. Pick two or three moments, each 8 to 25 seconds long, inside one shot. 6. Cut each one on the pauses, crop it vertical, and burn the disclosure on. 7. Write a caption per platform, with AD after the first sentence. 8. Put the clips in front of me to approve.

Steps one through seven take a couple of minutes per video. Step eight takes a couple of seconds longer than the clip itself: I watch it and proofread the captions. On the first afternoon I ran it, 11 videos turned into 33 clips.

Who should clip their YouTube videos this way?

It’s for you if you have a pile of long videos where you talk, and you’ve never cut any of them into vertical clips because editing feels like a second job (it is). Reviews, tutorials, vlogs, anything with a person speaking and a transcript to read.

Nothing in this system posts anything. It cuts, it writes, it waits. I say yes or no to every clip, and the approved ones go out through Blotato, the same pipeline my blog posts already use.

Skip it if any of these is you:

• You film vertical already. Then you don’t have a cropping problem, you have a scheduling one.
• Your videos have no talking. The whole system reads the transcript to find the moments. Music-only footage gives it nothing to read.
• You want the clips silent with a trending sound on top. These keep my voice on purpose. My app demos are the silent ones, and the two don’t mix.
• You want it to have taste. It picks well and it picks fast, but I still watch every clip, and I reject about a quarter of them.

My sister does videos occasionally too, and hers go through the same system. The only difference is the disclosure wording, which matches how she got the product.

What do you need to turn a YouTube video into Shorts?

Almost everything here is free, and the paid piece is the one doing the thinking:

• Claude Code. Paid. It reads the transcript, looks at the frames, picks the moments, and runs everything below. It’s the piece that makes “clip this video” mean something.
• A skill. One file that tells Claude Code exactly how to cut, in what layout, with what disclosure. Free. The generator at the bottom writes yours.
• yt-dlp. Free. Downloads the video and its captions from YouTube.
• ffmpeg. Free. Does the cutting, cropping, and stacking.
• Python with Pillow. Free. Draws the disclosure badge and pulls the frames for the contact sheet. My ffmpeg build has no text filter, so the badge is a small image laid over the video instead.
• Somewhere to approve clips. Optional. I built a one-page review screen that plays each clip and lets me edit captions and pick accounts. A folder and a text file work too, just slower.
• Blotato. Optional, paid. What I already post through, so approved clips join the same queue as everything else.

Nothing in that list is video editing software. I never open one. If the skill part is new to you, I wrote up what a Claude Code skill is and how to make one, and this one is the same shape with different rules.

How does a video get cut into Shorts without editing?

By cutting on pauses and picking the layout per clip. Both rules came from getting it wrong first.

Cut on pauses, never on caption timestamps. Cutting at the caption timestamps clips words. The system finds every gap in my speech instead, moves the start of a clip back to the end of the pause before the first word, and moves the end forward to the start of the pause after the last word. It’s allowed to drag a cut up to one second to reach a pause. Any further and a clip meant to start on the verdict starts on “my frother needs batteries.” When no pause is in reach, it cuts where the words are, fades the audio in over a tenth of a second, and tells me which end it did that to.

Pick the layout per clip. A 16:9 video loses two thirds of its width to become vertical, and where that slice comes from matters more than anything else here. Two layouts:

	Full frame	Stacked
The slice	One 608 pixel wide window from the 1920 wide source	Two 1215 pixel wide windows, each scaled to half a phone screen
What it fits	One subject, my face or the product	My face on top, the product on the bottom
How zoomed	Tight. A head and shoulders filling a phone	Half the zoom of full frame
When I use it	Product-only footage with no face in it	Almost every talking clip now

The first samples were full frame, and they were good, but it was so zoomed on my face. Stacked crops twice as wide and stacks the halves, so I read as a person in a room and the product gets its own half underneath. It tends to frame it all better. The bottom half is a product shot from elsewhere in the same video, and if it’s shorter than the talking half it plays forward then backward so the loop never jumps.

Where I sit in the frame. I sit to the right, around 62 percent of the way across, so a center crop cuts my face in half. The system doesn’t assume the center, and it doesn’t work it out from motion either, because motion tracks my hands and the product. It looks at a real frame and picks by eye, once per video.

The flagged word screen. Platforms transcribe the audio and read the text on screen, and certain words get a video buried with no notice. The first video I ever tried this on was a sleep hot chocolate called KillSwitch. I say the name out loud in the first four seconds, it’s printed on the tub, and the title card reads like a joke about dying. The transcript screen caught the spoken part. It can’t see packaging, so the contact sheet caught the rest, and the bottom-half footage had to come from the one clean shot in the video, the frothing on the wood table.

The disclosure. This one I caught. In a review, I say “sent to me to try” once, in the first few seconds, and a clip cut from minute two never hears it. Two of the three clips from my first run had the disclosure in neither the audio nor the picture. So now every clip gets the words burned on for its whole length, near the top, because the bottom quarter of a phone screen is covered by the platform’s own buttons. Clips where I say it out loud get the badge too, because people watch without sound. AD also goes in every caption, right after the first sentence. The wording follows how the product arrived, so my sister’s clips read “gifted product” instead.

Get the generator

The generator writes the part you’d otherwise learn by ruining a few clips, which is the rules. The ffmpeg commands are the easy part, and they’re included anyway.

You tell it what’s in your videos, which decides the layout. Where you sit in the frame, which decides the crop. What has to be disclosed, which decides the badge and the captions. Any words that must never appear in a clip, on top of the usual list. How many clips per video, and where they’ll go. It writes a skill for Claude Code that does the whole thing, cutting included, or a prompt for a browser chat that plans the cuts from a pasted transcript and hands you the commands to run.

Change two things in what it gives you: the crop, after you’ve looked at one real frame of your own footage, and the clip count, once you’ve seen how much of a two minute video is worth posting. For me it’s three.

What it doesn’t do

The crop is fixed for the whole clip, and that’s where most of my rejects come from. A clip that spans a close-up and a wide shot lands fine on one and badly on the other, and I’ve had the product end up half in frame more than once. The fix is keeping every clip inside one shot. The contact sheet shows where shots change, but on a couple of the early ones nobody looked.

The flagged word screen only reads the transcript. Packaging, labels, and title cards are invisible to it, and platforms read all three. Someone still has to look at the frames.

It can’t tell whether a moment is good. It can tell a line is complete, sits inside one shot, and has nothing from the flagged list in it. Whether anyone wants to watch it is my call, and I make it by watching every clip. About a quarter of the first batch got rejected, all for framing, none for the moment it picked.

The first frame of a clip isn’t always pretty. Snapping the cut to a pause fixes where the audio starts and does nothing for what my face is doing at that exact frame. TikTok lets me pick a cover frame at post time, so I haven’t chased it.

And it needs captions to exist. A video with none gives it nothing to read, and it stops at step one and says so.

Frequently asked questions

Try it yourself

Pick one long video where you talk, open its transcript, and find three sentences a stranger would understand with no setup. Write down the timestamp of each. That’s a clip plan, and it took five minutes.

Then run the generator above, look at one real frame before you trust the crop it suggests, and cut the first clip. Judge it on one thing: whether you’d have made it at all without the system. I’d been sitting on a thousand videos, and my answer was no.

Related reading

• Turn Blog Posts Into Social Media Posts Automatically With n8n
• What a Claude Code Skill Is, and How to Make One
• My Claude Code Setup: VS Code and a Mess of Folders

Subscribe if you haven’t already. I’ve got a long list of things I’ve built and never written up, and I’m working through it, so there’s more of this coming. In the meantime, go forth and automate the boring stuff.