Change aspect before you caption

Change the aspect, then burn captions. Change a video’s aspect ratio (edit_video) reframes the file. Burned-in captions are pixels. Type placed for 16:9 lives in a lower third that becomes the middle of a 9:16, or falls into the cut, or sits under TikTok chrome. Crop first. Caption the frame you will ship. Do not recaption a landscape burn and hope the platform ignores the extra canvas.

This page is the order: reframe, then type. It is not the safe-area chrome checklist, and it is not “generate 9:16 when you still can.”

Captions are a layout on a specific canvas

A 1920×1080 burn assumes a wide window. Karaoke hugs a lower third that is still above a YouTube control bar. A 1080×1920 burn assumes a tall window and a safe area above the username stack and inside the right-hand rail. Those two layouts are not a scale. Stretching the video does not move the words to the right place. Cropping the video moves or deletes the pixels they occupy.

So the sequence is:

1. Lock picture.
2. Reframe to the ship ratio (or export one file per ratio from an uncaptioned master).
3. Preview chrome if the destination is TikTok, Reels, or Shorts — including YouTube’s Shorts AI overlay on photoreal files (YouTube Blog, 27 May 2026): a chip on the video, not a line under the player.
4. Burn captions on that file.

Change video aspect ratio is the general tool: 9:16, 1:1, 16:9 in plain English to the agent. preview:true is a free 480p render, per-user cooldown. Final export is one editor_export. Clips must be HTTPS URLs.

Specialized doors if you already know the target: make-video-vertical, resize-video-for-tiktok, crop-video-to-square. Same rule. Aspect first. Type second.

Versely captions: add_veed_captions is 165 language codes, 21 BASIC looks at 1× and 9 DYNAMIC looks at 2×. Caption Studio has 45 CSS presets, 9 groups × 5 variants. None of those looks know you are about to punch the frame. They typeset the canvas you give them.

Sidecar SRT/VTT is not this burn. Sidecar timing can survive a crop better than pixels, but cue positions in some players are still canvas-relative. If you needed a sidecar, transcribe after picture lock and build cues for the file you upload. The burn is the short-form object.

What goes wrong if you caption first

Typical failures, all from one reversed order:

• 16:9 → 9:16 after a burn. The lower-third hook becomes a belt across the product, or it is cropped off with the bottom of the frame. The right-hand safe area never existed on the landscape file, so Reels chrome now eats a name you placed at the edge.
• 9:16 → 1:1 after a burn. Feed square eats the lower safe-area karaoke. The first word of the offer was the one sitting on the username stack’s old home, now the middle of a pack shot.
• 9:16 → 16:9 after a burn. You pad or crop. Padded bars make the captions look like a postage stamp. A tighter crop clips top and bottom type. Neither is a YouTube-ready lower third.
• Re-export at a new ratio with the same caption job settings. The engine may re-transcribe, but if you composite old burned pixels, you get two layers of type.

Fix: keep an uncaptioned master in the generate’s native ratio. Reframe copies from that. Caption each output. Disk is cheaper than a recaption plus a legal review of a clipped disclaimer.

If you already burned and then cropped, you cannot “move” the type. It is in the picture. Go back to the uncaptioned file. If you do not have one, you re-generate or you live with a crop that carefully avoids the old words — a composition driven by a mistake.

Reframe is a crop or a pad, not a restage

edit_video will not invent the top of a 9:16 that 16:9 never rendered. It will not invent the sides of a 16:9 that 9:16 never rendered. Preview the punch. If the SKU dies, regenerate at the ship ratio with the still cropped first (I2V) or the aspect set on T2V. Captioning cannot restore a missing label.

Multiple destinations in one week:

• Reels / TikTok / Shorts: 9:16, captioned for vertical chrome.
• Instagram feed: 1:1 or 4:5, captioned for a shorter canvas.
• YouTube long-form: 16:9, captioned for a lower third, sidecar if the player should toggle.

That is two or three caption jobs, not one burn scaled. The uncaptioned master is the shared parent. Add captions to video on each child.

Do not caption “while the preview renders.” Preview is 480p and for framing. Type at 480p that you then blow up is a different softness. Caption the export you will upload.

Uncaptioned master, then children

Procedure:

1. Generate (or shoot) in the primary ratio if you still can. If you cannot, pick the file you have.
2. Export or keep a copy with no burned type.
3. edit_video to each ship ratio. Preview 480p. Confirm face, pack, legal plate.
4. Caption each ratio. Check safe area on a phone.
5. Never recaption by cropping a captioned file.

Legal lines and SKUs in captions are the expensive miss. A disclaimer that sat in a 16:9 lower third can lose its last clause in a 9:16 punch, or sit under Reels chrome so the required text is not actually on screen. Recaptioning after a crop is not polish. It is how you notice the line is gone. Preview the reframe with the picture of where type will go — a mental block in the safe area — before you burn. If that block covers the pack, the crop is wrong or the caption must move up, not “scale.”

DYNAMIC looks at 2× (glass, whisper, glide, and the rest of the nine) are worse to redo than BASIC at 1×. Pick the look on the canvas you will ship. A karaoke highlight timed to a landscape lower third will not reflow onto a vertical safe area because you cropped. The word clock is still the audio; the placement is the canvas.

If you need a seat to reframe the keeper and then burn type, sign up. Bring the uncaptioned master.

FAQ

Can I preview the reframe before I pay?

Yes. preview:true is a free 480p render, subject to a per-user cooldown. Use it to see whether the SKU survives. Caption after the charged export of the ratio you will ship.

Do sidecar SRT files need the same order?

Build cues against the file you upload. A crop changes which words sit on which pixels even if timestamps still match the audio. Transcribe the locked mix; author the sidecar for that canvas.

What if I already burned captions on 16:9?

Do not punch that file for TikTok. Return to an uncaptioned copy, reframe, recaption. If no uncaptioned copy exists, regenerate or accept a vertical crop that avoids the old type — usually the worse picture.