Captions and subtitles

Prompt

Add captions to [CompId].

Source audio: [VO file]. Script: [script text or "transcribe"]. Languages: [languages]. Style: [word-by-word highlight | line captions | .srt only]. Formats: [16:9 | 9:16 | 1:1].

1. Timing: get word-level timestamps from the audio (transcription), then correct the words against the script so the spelling is exact. Report any place where the speech differs from the script.
2. Segmenting: break into caption pages of at most [2] lines and about [32] characters per line for 16:9 ([20 to 24] for 9:16). Break at natural phrase boundaries, never between an article and its noun or a number and its unit. Each page on screen at least 1 s; extend the last word by 0.2 to 0.3 s.
3. Style (burned in): one type family from the design system, high contrast (a soft dark box or strong shadow when the background is busy), positioned inside the safe area and never over the focal element. For word-by-word highlight, the active word changes colour or weight only; no bouncing words unless the brief asks for it. No em or en dashes as separators; use a comma or colon.
4. Check collisions: list the frames where captions overlap on-screen titles, lower thirds or UI and move or hide the caption for that beat.
5. Deliver .srt for every language, even if captions are burned in.
6. Render stills at the busiest frames and the longest caption page, in each format, and inspect them.