Short-form captions
Use the active edit and its visual language. Captions are language first: accuracy, sync and reading comfort matter more than word effects.
- Transcribe actual speech and keep corrected timed data in SRT, VTT or the editor's editable format. Check names, brands, numbers, units, negations and qualifiers against the audio. Never silently substitute an approved script for different spoken words. When a take contains restarts, transcribe pause-split chunks separately so a repeated line is not merged into one clean sentence; the editor's local tools show how.
- Segment by meaning, breath and syntax, normally one or two lines. Keep related words together and avoid orphaned articles/prepositions. Change line breaks without rewriting speech. Hold phrases long enough to read without showing the next thought early; re-time after any speech edit.
- Prefer a stable phrase caption with selective emphasis. Use karaoke only when requested or supported by the creator's established treatment. Do not reserve a large empty bottom band or add a second burned-in transcript layer.
- Follow the existing font and contrast treatment. Start around 46–68 px at 1080-wide resolution, then inspect the rendered size rather than trusting an editor's font units. Use a panel, outline or shadow only as needed.
- Follow the adaptive editor's safe-area guidance when installed. Center main captions on the whole canvas; check the complete text, backing and shadow, not only the anchor. Keep the face and demonstrated detail unobstructed. If using this skill alone, measure against the intended platform preview and check the longest captions at phone size.
- Watch an encoded proof with sound and silently when playback is available. Check phrase boundaries, changing backgrounds, speaker changes, two-line bounds and the ending. Report a missing listening pass explicitly.
Label translations and paraphrased summaries when material; distinguish them from verbatim captions. Deliver the corrected timed source with the revised video or subtitle file requested. Preserve previous versions and the rest of an approved edit. If the creator will add captions in their own app, provide clean footage and subtitle data as needed instead of baking them in twice.