Reel auto-captions misspell names and brands: fix with script_text

Send the correct text as script_text when you burn captions: Sume keeps the speech-to-text word timings as the source of truth for time and aligns your spelling to them. Machine captions are weakest on names, slang, idioms and technical terms, which is exactly where a brand cannot afford a typo.

GrowthViral's review of Instagram's bilingual captions says the same of automated captions in general: names, slang, idioms and technical terms need manual review (GrowthViral, read 2026-10-07). A burned-in caption cannot be corrected after posting, so the fix belongs before the render.

The call

On video captions, script_text is optional. The clip must have audible speech and be a public HTTPS URL. Sume still transcribes the audio to get timings and then lays your text over them.

curl -X POST https://api.sume.com/v1/video-captions \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: launch-reel-script-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/example/clean.mp4",
    "style": "punch",
    "script_text": "Meet the Sume developer platform."
  }'

What can go wrong

Alignment can fail if your script differs a lot from what was spoken. The docs name two typed errors and the next step for both.

Three ways to correct, by situation

• You wrote the script and read it: send script_text on the first burn.
• You ad-libbed and only a few words are wrong: burn once, then restyle with source_caption_id and send words with the corrected text. Speech-to-text does not run again.
• A silent or music-only clip: skip speech entirely and send cues with text, start and end.

What to check before posting

Read the burned text against your script once, out loud. Names and numbers first. Only one of script_text, words, cues and segments is accepted per request, and each accepted job is $0.20 for videos of up to 60 seconds, so three corrected passes cost $0.60. A free check is cheaper: read a transcript from video inspect ($0.01 per audio minute plus compute) before you burn.

Write the script to match the speech

Alignment works best when the script is close to what was said. Write it as you speak, including contractions, and leave out stage directions. If you read a script on camera and changed a few words, edit the script to match the take rather than the other way around. A small mismatch is usually fine; a large one raises script_alignment_mismatch.

If a take has a long ad-lib in the middle, burn the first part with the script and the second part from speech-to-text, as two clips joined later. That is more work, but it prevents a failed alignment over a sentence you never wrote.

Keep a name list

Keep a short list of the names, brands and product terms you use most, and paste it into the top of every script. It is also a handy review checklist: before you post, search the burned captions for each name and confirm the spelling on screen.