Restyle holiday ad captions for an A/B test without re-transcribing

To test caption looks on one holiday ad, burn it once, then restyle it by sending the first job's id as source_caption_id. Sume reuses the word timings it already has and does not run speech-to-text a second time. Each restyle is still a render, so each costs $0.20.

This is the pattern in the video captions docs: the second request carries source_caption_id and a style, and no video_url.

The test plan

Pick two or three styles from the list the docs give: slam, punch, tiktok-green and, for Korean audiences, korean-ad. Run the first with the video URL, note the caption id in the result, and then send one restyle per additional look.

What to hold constant

Change only the caption style between variants. Keep the cut, the audio and the first frame identical, otherwise you cannot tell what moved the result. Upload each version under a name that includes the style so the report lines up.

If the transcript has a wrong word, send corrected words with the restyle. The docs say to use that field only to correct the text.

Limits

Standalone caption jobs are priced for videos of up to 60 seconds. A holiday ad of 15 or 30 seconds fits, and a 90-second cut does not.