Talking-head post-production: cut flubs, add captions, publish

Talking-head flubs have fixed shapes: long pauses between prompter lines, retakes after a stumble, downward glances at the script. Trim segment by segment (keep the good take, cut the rest) rather than fine-editing — one card per sentence, cards back to back, and the rhythm lands clean. Rather keep half a second of breath than one wasted sentence.

Order matters: cut the picture first (flubs and pauses), add captions after — captions placed before cutting stop matching the picture and the rework is expensive. This is the general rule: captions come late in post.

Captions matter more for talking heads than almost any other content: feed viewers watch muted and understand through text — caption coverage directly drives completion. Auto-recognition lays the base; humans proofread brand names, jargon, and numbers, which are exactly where ASR errs most and where viewers notice.

Keep caption styling plain: white with a dark edge, bottom center, no more than about fifteen characters per line. Talking heads are watch-the-face content; flashy captions steal the show. Once verified, burn them in — platform distribution ignores external subtitle tracks, and burning is the only reliable path.

The audio chain: denoise the voice (room noise drives viewers away faster than visual flaws) → lay BGM underneath (start at eighty percent voice, twenty percent music, listen until every word is clear) → one full loudness pass. Pick instrumental BGM so nothing fights the voice.

Compress to the target platform spec, then watch the whole thing once — caption typos, audio clipping, and dropped frames are all catchable in one full pass. The feed gives you three seconds; polishing those three seconds of picture and sound matters more than any parameter.