Narrate a 3-minute Short: Sume TTS, word timings and 60 s caption cuts
YouTube allows three-minute Shorts. A plan with Sume: one TTS request under 20,000 characters, word timings, and caption jobs in pieces of 60 seconds or less.
YouTube Help says Shorts can be up to three minutes (read 2026-10-03). On Sume, narrate one with a single tts_create request (the transcript limit is 20,000 characters, far above what three minutes of speech needs), ask for timestamps.words, and cut the captions into pieces of 60 seconds or less, because Sume's caption job covers videos up to 60 seconds. At $0.0475 per 1,000 characters, a 3,000-character script costs $0.1425 before the minimum applies.
The plan
The listed rates are in the Sume catalog, and you should confirm them live with GET /v1/catalog.
| Step | Surface | Limit or price |
|---|---|---|
| Voiceover | tts_create | Up to 20,000 characters; $0.0475 per 1,000 characters |
| Word timings | timestamps.words true | Returns words[] with start and end |
| Captions | POST /v1/video-captions | Up to 60 seconds per video, $0.20 each |
| Join | Timeline 1.0 | $0.10 per output minute, rounded up |
Cutting at 60 seconds
Split the words[] list at a sentence end near each 60-second mark, subtract the piece's start time from every start and end, and send each piece as words. Three pieces make three caption jobs. Then join the captioned pieces in Timeline. See the captions guide for the word limits.
Sources
More in Use cases
- Neighborhood guide video for real estate agents: photos and music
A 30-second neighborhood guide built from your own photos of local spots: stills in Timeline 1.0, one caption cue per spot, and a soundtrack bed.
- One MP4 for Instagram Reels, Facebook Reels, Stories and Threads
One 9:16 export can pass Instagram Reels and Stories, Facebook Reels and Page stories, and Threads: 60 s, 100 MB, 24-60 fps. The shared spec and Sume settings.
- Open house promo reel: date, time and address over listing photos
Make a vertical open house reel from listing photos: a silent Timeline 1.0 render, then burn date, time and address as caption cues. Rates and limits included.
- Partnership ad code or account-level permission: which to ask for
Meta partnership ads come in two permission types. How content-level codes and account-level access differ, and what to log when the clips are AI-made.
Written by Sume