Transcript captions

Make each cue effortless to read at the speed and scale of the final video. Preserve semantic subtitles when accessibility or localization matters; use Text+ when typography and motion are part of the edit.

Choose the representation

• Use native subtitles for searchable, editable, exportable language tracks.
• Use designed Text+ captions for branded typography, per-cue composition, emphasis, or motion.
• Use both when the video needs a visible design and a semantic sidecar/localization source.

For local execution, read cutagent-captions, cutagent-fusion; check the installed SDK declarations, public CutAgent CLI command cards, and live capabilities for exact operations.

Segment for meaning

Work from a fresh transcript of the actual target timeline. Break on speaker changes, completed thoughts, clause boundaries, and meaningful pauses. Keep tightly connected grammar together and avoid orphaning one word. Every spoken token should appear once unless the user explicitly requests editorial paraphrase.

Choose cue length from the frame size, type scale, language, speech pace, and design. Prefer one line for fast social captions; use two lines only when the composition and reading rhythm benefit. Give short cues enough screen time to register and split dense cues before shrinking the type into illegibility.

Design the system