Convert a long cut into a Short

Keep the moments, stitch them, then reframe to 9:16. Convert a video to Shorts is two decisions: which seconds of the long cut are the ad, and which window those seconds must fill. cut_video keeps ranges. edit_video sets 9:16. Captions belong on the Short. A burn on the 12-minute master will sit in the wrong third of a phone.

This is a cutdown of a file you already have. It is not “generate 9:16 first,” and it is not a hook pack of new generates.

Pick keep-ranges, not a delete list

cut_video segments are KEEP ranges, in seconds, half-open [start, end). One range is a trim. Several ranges concatenate in the order you give them. The tool does not rank “best moments.” You still watch the long cut.

Procedure:

1. Watch the long file once for the claim. Write the sentence the Short must say even on mute.
2. Watch again with a pad and timestamps. Mark proof (the verb, the pack, the face) and the line that carries the offer.
3. List keep-ranges. A Short is usually 15–45 seconds of this job, not the whole webinar. Two or three ranges beat eight.
4. Order them as a Short, not as a documentary. Hook first even if that moment happened at 14:02. The concatenate order is the edit.
5. Run cut_video. Inspect each seam. A hard cut from applause to a pack is fine. A hard cut that lands on a blink needs a different in-point or a cover.

What to keep:

• A first-second picture that is already a claim: result, mess, pack, face. Not a logo hold. Not “welcome back.”
• One verb. Unscrews, pours, clicks. Not four verbs from four chapters.
• The line you would caption. If the proof is silent B-roll, you will add voiceover later; still keep picture that matches the line.

What not to keep:

• Intros, recaps, “as I said earlier.”
• A second example that repeats the first. One proof.
• Lower-thirds and slide decks burned for 16:9. They will die in the vertical crop.
• Dead air you could have trimmed. If the keep-range includes a 4-second nod, you built a slow Short.

You specify the ranges. The tool will not auto-detect virality. analyze_video can dump beats and on-screen text if you need a map of a 40-minute file, but it still does not pick the ad. Do not outsource the keep list to a model that has not seen your offer.

If the long cut is already several clips on a timeline you do not have, export a flattened master first. cut_video works on one source URL.

Reframe after the cut, not before

Do not 9:16 the 12-minute file and then hunt. You will crop a two-shot for an hour and then throw most of it away. Cut to the keepers in the original window, then reframe that short file.

Make a video vertical is the 9:16 render via edit_video. Tell the agent whether to crop tight on the subject or pad. Padding on a Short is black or blur pillars. Crop is full-screen and discards left and right of a 16:9 master. Pick. Letterboxed Shorts read as unfinished.

Geometry: a centre crop of 16:9 to 9:16 keeps about a third of the width. Two-shots lose a person. Hands on a table disappear. A slide with a sentence in the lower third becomes a fragment. If the keep-range is a wide two-shot, either pick a different range or accept pad — or do not convert this moment. Regenerating that beat at 9:16 is a new job, not this conversion.

Keyframe the crop when the subject travels. A pack that starts left and exits right needs a moving window, or you crop it out in the second half. Preview at phone size. preview:true is a free 480p check, per-user cooldown. Final export is one editor_export.

YouTube Shorts is 9:16. TikTok is 9:16. Reels is 9:16. The player matches. Chrome does not. Recaption per destination if the burn sits on a username stack. YouTube’s photoreal AI overlay sits on the Short (YouTube Blog, 27 May 2026), not under the player the way long-form labels do. Leave room at the top-right of the vertical frame if the file is photoreal / meaningfully AI-altered.

Do not upscale a thin strip and call it 4K. You cannot invent the pixels 16:9 never rendered above and below. If the long master was 1080p 16:9, the Short is a crop of that.

Caption the Short, not the long master

Burned-in captions are pixels. A transcript of the 12-minute file will include chapters you deleted. A burn placed for 16:9 sits in the middle of 9:16 or under chrome.

Sequence:

1. Cut keep-ranges.
2. Reframe to 9:16. Preview the crop.
3. Audio lock. If you laid a new VO on the Short, replace or mix on this file.
4. Transcribe and burn. Add captions to a video: 165 language codes, 21 BASIC at 1×, 9 DYNAMIC at 2×. Language must match what is now spoken. It does not translate.
5. Watch muted on a phone. First line is the hook test. If it sits under the username, move the block up and burn again.
6. Overlay anything nobody said: URL, price, “Part 1.” That is not a caption track.

Sidecar SRT from the long master is a starting transcript, not a burn. Timing is on the old clock. After a concatenate of two ranges, every cue after the first seam is wrong. Recue on the Short. What captions cost across a month of clips is why you do not caption the master “so we have it” and then recaption the child.

Mute-autoplay is the environment. Kapwing’s 2026 roundup still treats on-screen words as the lever when the speaker is off. The long-form viewer might have had headphones. The Shorts viewer often does not.

One Short is one job

A 10-minute podcast is not one Short. It is several Shorts if several claims exist. Convert is not “compress the episode.” Each output gets:

• One claim.
• One proof range (or two that stitch).
• One 9:16 crop decision.
• One caption pass.

Do not concatenate six highlights into a 90-second “best of” and call it a Short. That is a trailer. Trailers have a different first second.

If you need the same keep-range in 1:1 for feed, crop square from the uncaptioned 9:16 child, then caption the square separately. Type does not survive a second punch.

Article 50 (applies 2 August 2026) still wants a deployer label at first exposure on photoreal AI that would pass as real. A cutdown does not wash that off. YouTube’s overlay is not a substitute for a disclosure your jurisdiction requires. Labels alone do not decide monetization (YouTube Blog, 27 May 2026); inauthentic-content policy is a different door.

Convert the long cut in Versely: keep-ranges, then 9:16, then captions on the Short. Do not burn the webinar and crop it later.

FAQ

Does it automatically find the best clips?

No. You pass keep-ranges to cut_video. The tool cuts precisely. Ranking moments is still yours.

Can I stitch non-adjacent moments?

Yes. Multiple segments concatenate in list order. Hook at 14:02 can play before proof at 3:10 if you list it first.

Should I caption the long master first?

No. Caption the Short after the crop. Long-master burns are the wrong canvas and the wrong transcript.