How to Make Shorts From a YouTube Video: The Complete 2026 Walkthrough

There is a version of this task that takes an afternoon and a version that takes two minutes, and the difference has almost nothing to do with software. It comes down to whether you know which decisions matter. Four of them do. The rest is button pressing that any tool will handle for you.

I have cut a lot of these, mostly from podcasts and long interviews, and the pattern is consistent. People spend twenty minutes fiddling with caption fonts and about forty seconds choosing the moment, when the moment is the entire product. A perfectly styled clip of a boring thirty seconds is a boring clip. A badly styled clip of a genuinely interesting thirty seconds still travels.

So this walkthrough is weighted the way the work should be weighted. The technical steps are here in full, with real numbers, but the longest section is the one about choosing what to cut.

What a Short actually is, technically

Strip away the branding and a YouTube Short is a video that satisfies three conditions: it is vertical or square rather than landscape, it runs no longer than three minutes, and it was uploaded as a normal video that YouTube then classified into the Shorts feed. There is no separate Shorts file format. The upload path is the same.

The practical specification everybody uses is 1080 pixels wide by 1920 pixels tall, which is a 9 by 16 ratio, in an MP4 container with H.264 video and AAC audio. That is not a rule imposed by YouTube so much as a convention every short form platform converged on, which is convenient: the same file posts to Shorts, TikTok, Instagram Reels, and Facebook Reels without a second export.

The three minute maximum arrived in late 2024, up from sixty seconds. It is worth knowing that limit exists and then largely ignoring it, for reasons covered in the length guide. There is also a licensing catch buried in it: if you use a track from YouTube's licensed music library, your Short is still capped at sixty seconds. Original audio and royalty free tracks are what unlock the full three minutes.

Step one: choose a source worth clipping

Not every long video contains a Short. Recorded talks, podcasts, interviews, tutorials, product demos, and streams are dense with them because somebody is saying things. Ambient content, B roll heavy travel films, and anything where the value is cumulative rather than momentary tend not to be.

The test I use before opening anything: can I name the moment out loud? If I can say "the bit where she explains why the first version failed" then there is a clip in there. If the best I can do is "the middle section is good" then there probably is not, and I am about to waste twenty minutes finding that out the hard way.

One practical habit that saves more time than any tool: note timestamps while you record or during your first listen back. Ten seconds of typing during the conversation removes twenty minutes of scrubbing later. Podcasters who clip consistently almost all do this. Podcasters who stopped clipping by episode six almost all did not.

Step two: find the moment, the part everyone rushes

This is the section that matters. A moment that works as a Short has a specific shape, and once you see the shape you cannot unsee it.

It has to stand alone

The person watching has not seen the rest of the video and has no loyalty to you. They arrived by accident. If the clip requires knowing who these people are, what was said five minutes ago, or what the running joke means, it is a clip for your existing audience. That has value, but it is not discovery, and discovery is what Shorts are for.

It opens on a hook and closes on a payoff

The most reliable structure is a claim or a question that creates a small gap in the viewer's understanding, followed by the thing that closes the gap. "Everyone tells you to post daily. That is the worst advice for a channel under a thousand subscribers, and here is why." The gap opens in the first three seconds and closes before the clip ends. That is a Short.

What does not work is a clip that opens the gap and then runs out of time. People feel interrupted, and being interrupted reads as a bad video rather than a cliffhanger. If the payoff needs another twenty seconds, take the twenty seconds.

Something has to be happening visually

Short form is a visual feed even when the content is conversational. Moments where somebody leans in, gestures, laughs, or reacts consistently outperform moments where two people sit still saying accurate things. This feels unfair if you care about substance, and it is true regardless of how you feel about it.

Where AI assistance actually helps

Reading a transcript for candidate moments is exactly the kind of tedious pattern matching software is good at. A tool that scans the whole transcript and hands you ten ranked candidates saves the scrubbing, and on a two hour episode that is a real saving.

What it cannot do is know your audience. The ranked list is a starting shortlist, not a verdict. Every time I have posted the top ranked suggestion without checking it, the clip was fine and forgettable. The good ones came from the list at position four or seven, where the model saw something interesting and I knew why it mattered.

Step three: cut it so it opens and closes cleanly

Once you have the moment, the cut itself is mechanical, with two rules.

Cut the run up. Almost every good moment is preceded by three or four seconds of throat clearing: "yeah, so, I think the thing is". Delete it. Start on the first word that carries meaning. This single edit does more for retention than any caption style you will ever pick.

Do not cut mid word. Frame accurate trimming exists precisely so a clip does not start halfway through a syllable. Any decent tool snaps to sensible boundaries, but check the first and last half second by ear before exporting, because a clipped consonant at the start sounds like a broken file rather than a stylistic choice.

A useful trick for the ending: cut on the beat after the last important word, not immediately on it. Roughly a third of a second of air lets the sentence land. Cutting instantly on the final syllable feels rushed in a way people notice without being able to say why.

Step four: go vertical without decapitating anyone

Here is the part that separates tools. Your source is almost certainly 1920 by 1080. The Short is 1080 by 1920. Same height, but the width goes from 1920 to 1080, which means about 44 percent of the original width survives and 56 percent has to go. On typical framing where the subject sits off centre, that is enough to remove a face.

There are three honest ways to handle it, and they are not equal.

The failure case worth understanding is a wide two shot, the standard podcast setup with a guest on the left and a host on the right. The geometric centre of that frame is the gap between them. A fixed centre crop on that footage produces a vertical video of a table with two shoulders at the edges. It happens constantly with cheap converters, and it is why subject aware framing is not a luxury feature.

Good implementations also know when to give up. If the tool cannot find a subject with confidence, staying centred is better than committing to a wrong guess, because a centred crop is merely imperfect while a confidently wrong crop is unusable.

Step five: captions that read at a glance

Most short form viewing starts muted. Captions are therefore not an accessibility afterthought, they are the primary channel for anything spoken. Three things matter and the rest is taste.

• Timing. Word by word, synced to the audio. Big blocks of text that sit still for six seconds are worse than nothing, because the eye finishes reading and the brain disengages.
• Contrast. Heavy weight type with an outline or a shadow. Thin elegant fonts disappear against real footage the moment the background gets busy.
• Position. Keep text out of the bottom fifth and top eighth of the frame, where platform interface elements sit. On Shorts that is the title, the channel handle, and the description bar; on TikTok it is worse. Text that lands under the interface looks like a mistake by the creator, not by the platform.

Read the captions before you export. Speech recognition is strong on ordinary sentences and weak exactly where it hurts: names, brands, jargon, and heavy accents. Getting a guest's name wrong in burned in text is not fixable after posting.

Step six: export settings that never get rejected

There is very little to decide here, which is good, because this is the step where people waste time on settings that make no visible difference.

Two notes. Do not upscale a 720p source to 1080p thinking it helps; it inflates the file without adding detail. And do not change the frame rate unless you have a reason, because converting 30 to 60 or the reverse introduces judder that is subtle on a desktop and obvious on a phone.

The full specification, including safe zones and what happens when you upload the wrong shape, is in the size and specs reference.

Step seven: posting it so it gets a fair chance

Upload it like a normal video. YouTube classifies anything vertical and under three minutes into the Shorts feed automatically, so there is no separate button and no hashtag required. The old advice about adding a hashtag to force classification is out of date.

What still matters:

• The title does work. It appears over the video in the feed and it is the only text a scroller reads before deciding. Write the hook, not a description of the content.
• The first second is the thumbnail. Shorts autoplay, so whatever frame opens the clip is doing the job a thumbnail does elsewhere. A clip that opens on a blurry transition wastes it.
• Link back in the description, not in the clip. Telling people to watch the full episode inside a thirty second video costs you retention on the thing you are promoting. The link catches the people who already want it.
• Post to all three platforms. The file is identical for Shorts, TikTok and Reels. Just export clean copies rather than downloading your own TikTok with its watermark burned in, which every platform quietly deprioritises.

The mistakes that quietly kill clips

None of these produce an error message. They just produce clips that do nothing, which is harder to diagnose.

• Starting three seconds too early. The single most common one. Those three seconds of "so, yeah, I think" are where most of your audience leaves.
• Posting a clip that needed context. It made you laugh because you watched the whole episode. A stranger sees two people laughing at nothing.
• Letting the crop cut somebody in half. Preview the vertical frame before exporting. Every single time.
• Captions under the interface. Looks fine in your editor, unreadable in the actual feed.
• Uploading a landscape video and hoping. It will not be classified as a Short. It will sit in your normal uploads with black bars, getting no distribution from either feed.
• Chasing volume over selection. Thirty mediocre clips from one video will not outperform three good ones, and they train the algorithm that your clips are worth skipping.

Questions people ask

The part worth remembering

The technical steps in this guide will be handled for you by any competent tool, including ours. They are worth understanding so you can tell when a tool has done them badly, which happens more than you would expect, especially with the vertical crop.

The part no tool does for you is the second step. Choosing a moment that stands on its own, opens on the hook, and closes on the payoff is the whole job. Everything after it is production. Spend your time accordingly and the clips will look after themselves.