Podcast sponsor clip: restyle captions without a second transcription

To get one podcast sponsor clip in several caption looks, caption it once and then send source_caption_id with a new style instead of the video URL. Sume reuses the source video of that caption and the word timings it already has, so speech-to-text does not run a second time. The price does not change, because a restyle is still a render. That makes it a cheap way to test which look reads best before the holiday sponsor reads go out.

How the restyle request differs

The first request carries video_url and runs transcription. Every later look is a small body: the earlier caption id and the new style. If a word was heard wrong, send words with the request to correct the text; the docs say words are only needed for corrections on a restyle.

Steps

Plan the looks you want to compare before you start so you only transcribe once.

• Caption the sponsor clip with the style you expect to use most, and keep the returned caption id in your notes.
• For each other platform, POST a body with source_caption_id and a different style, using a new Idempotency-Key for each look.
• Remember that design overrides do not apply to the punch and tiktok-green styles, so tune colors on the other styles.
• Pick the winner by watching the burned files on a phone, since small text is where the styles differ the most.

curl -X POST https://api.sume.com/v1/video-captions \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: sponsor-clip-restyle-black-outline" \
  -d '{"source_caption_id": "YOUR_CAPTION_ID", "style": "black-outline"}'

What Sume does not do

A restyle changes the look and not the audio or the cut. It does not shorten the clip, and it does not check that the sponsor line or promo code is correct; if the transcription heard the code wrong, correct it with words and then look at the frames. The caption endpoint also needs a public video URL, so a private file must be hosted first.

If the brand name is spelled wrongly in the first pass, fix the cause with the script option described in the linked post and then restyle from the corrected caption, so you do not repeat the fix for every look.

A holiday workflow

Sponsor reads cluster in November and December, and the same 30 to 60 second clip often needs a feed version, a story version and a site version. Treat the first caption call as the expensive thinking step: pick the cleanest cut, check that the transcript has the sponsor name and any code spelled correctly, and only then fan out the looks.

Store the caption id next to the episode and sponsor in your own sheet. The docs describe the id as the handle for restyling, and they do not describe a way to list your earlier caption jobs by episode, so your own record is the practical index. Keep the style name you used too, so you can tell later which file was the black outline version and which was the tiktok-green one.