How to turn long videos into Shorts, Reels and TikToks with AI

A one-hour webinar usually holds five to ten moments that would work on their own as a short. Finding them by hand takes an editor most of a day. AI clipping does the first pass in minutes. This guide explains how the pipeline works, the current platform specs, and what a person still needs to check.

Turning a long video into shorts with AI means transcribing it, letting a model pick the strongest self-contained moments, cutting them on word boundaries, reframing each shot from 16:9 to vertical 9:16, and burning in captions that stay clear of each app's buttons. The output is a set of ready clips you review, edit and schedule.

How does AI clipping work, step by step?

AI clipping is a pipeline of five stages, and knowing them tells you where to look when a clip comes out wrong. This is how Tracee's clipper works; most tools follow a similar order.

1. Transcribe. Speech-to-text produces the transcript with a timestamp for every word. Everything after this works from the words, not the pixels.
2. Score moments. The transcript is split into windows, and each window is scored 0 to 100 on whether its first two seconds would stop someone who has never heard of you. The best windows go to a second pass that cuts the actual clips, writes a hook line and a title for each, and scores them again.
3. Cut cleanly. The best clips by score are kept, overlapping ones are dropped, and each start and end is moved to a word boundary so no clip begins mid-syllable.
4. Reframe to vertical. The clip is split at scene changes, and a vision model looks at frames from each shot to find faces and screens. Each shot then gets its own layout (next section).
5. Caption and hook. Word-timed captions are burned in a few words at a time, with the spoken word highlighted. A short hook line sits over the first seconds, placed where it covers no face and no caption.

In Tracee, a source can be up to two hours and 1 GB, and one job returns up to 12 clips, best first. Each clip comes with its score, hook, transcript and ready-made copy for TikTok, Instagram and YouTube Shorts.

How does reframing from 16:9 to 9:16 decide what stays in frame?

Reframing picks a layout per shot, because one fixed crop fails as soon as the camera cuts. Tracee's clipper uses four layouts, chosen from what the vision model saw in each shot:

The output is 1080 by 1920 pixels. Short shots between two that agree inherit their layout, which avoids a one-second jump to a different framing.

Reframing fails in predictable places: fast movement, two speakers who keep leaning in and out, and slides with small text. Those are the shots to watch on review.

What are the specs for YouTube Shorts, Reels and TikTok?

All three platforms take vertical 9:16 video, and 1080 by 1920 works everywhere. Length and safe zones differ. Organic limits change often and can vary by account, so check the upload screen when a number matters.

Sources, as of November 2026: YouTube's three-minute Shorts article; Meta's Instagram Reels ad specs; TikTok's in-feed ad specifications (updated June 2026); and PetaPixel's report on Instagram's 20-minute Reels camera (November 2025). The Meta and TikTok numbers come from ad specs, which are the most precise official figures each company publishes; organic uploads follow the same shape.

One YouTube rule affects clipping directly: "Any Short that is over one minute in duration with an active copyright claim of any type" is blocked globally, per YouTube's article. If your source has background music or third-party footage, keep clips under a minute.

What should you check by hand before posting a clip?

AI gets the first cut close, but five things still need a person, and they take two to three minutes per clip.

1. The cut points. Does the clip start on the claim and end after the point lands? Trim a second or two if the first words are a filler sentence.
2. Context. A sentence that's fine inside a 60-minute talk can read as a different claim on its own. Read the transcript of the clip as a stranger would.
3. Facts and names. Check numbers, product names and people's names in the captions. Speech-to-text gets brand names wrong more often than common words.
4. The hook line. It must be true to what the clip says. A hook that promises more than the clip delivers costs you trust and watch time.
5. Framing on every shot. Scrub through once and look for cut-off heads, unreadable slides, or a split layout on the wrong two people.

Also confirm you have the rights to clip the source: your own webinar is fine, a conference recording may belong to the organizer, and a guest on your podcast may expect to approve clips that feature them.

What does a weekly repurposing workflow look like?

One long video a week can feed every short-form channel for that week. A workflow that fits a small team:

1. Monday: Clip last week's webinar or podcast episode. Take the top five clips by score.
2. Monday: Review each with the checklist above. Trim, fix the hook, or re-make a clip with a different caption style.
3. Tuesday: Write one version per channel from the ready-made copy: TikTok and Instagram descriptions, a YouTube Shorts title. Add a link to the full video where the network allows it.
4. Tuesday: Schedule one clip a day across TikTok, Reels and Shorts. LinkedIn and X can take the same clips with a text-first caption.
5. Following Monday: Compare which clips held attention and brought clicks, and note the hook patterns that worked for the next round.

For the long video itself, generate YouTube titles, a description with chapters and thumbnails at the same time. The AI content calendar guide shows how to slot clips into a month plan.

How do you clip a video with Tracee?

Tracee clips from the Clips page, from the Tracee CLI, or through an AI agent connected over MCP, so you can ask Claude or ChatGPT to clip last week's webinar and draft the posts.

The source can be a video page on YouTube, Vimeo, TikTok, Instagram, X, LinkedIn, Loom and other sites, a Google Drive, Dropbox, OneDrive or Box share link, a direct file link, or a video already in your library. From the CLI:

tracee clips create --url https://youtu.be/<video_id> --caption-style hormozi
tracee clips get <job_id>
tracee posts create --body "<the clip's copy>" --channel tiktok --media <media_id>

After clipping you can trim a clip, change its hook, re-make captions, let a model add zooms and color accents on key words, dub a clip into another language such as Spanish or Japanese, or package the long video for YouTube with ten title options, a description with chapters and three thumbnails. Agent-drafted posts with the clips wait for approval by default. The developer guide to clipping lists every option.

How much clipping does each Tracee plan include?

Clipping is counted in minutes of source video per month. A 45-minute webinar uses 45 minutes, however many clips it produces. As of November 2026 (pricing):

Failed clips are refunded. Allowances reset every month, on yearly plans too, and unused minutes don't roll over. Beyond the allowance, clipping uses credits at 1 credit per minute, bought in packs that don't expire while your account is active.

Next step

Pick one recording from the last month that already did well as a long video, and clip it. Review the top five clips with the checklist above and schedule them over a week. Tracee's 7-day free trial includes clipping, so you can test one real webinar before deciding.