---
description: Tighten a talking-head video without destroying natural speech rhythm
---

Use the `talking-head-edit` skill to perform a full expression edit by default: review the entire source, select complete takes, remove false starts and redundant retakes, preserve/reinsert unique information, then adjust pauses; use pauses-only scope only on an explicit user restriction; submit talking_head_editorial_plan before applying A-roll, retain its receipt, distinguish draft previews from content-reviewed timelines and pending output listening; review full-sentence context, word-level pauses, filler words, adjacent repetitions, and low-confidence delivery cues; stabilize A-roll; identify only exact visual cues that add visible information; keep openings, conclusions, opinions, emotions, and transitions on A-roll by default; let the topic and visible information determine total B-roll coverage; count A-roll islands under three seconds inside the same viewing run; reserve three seconds of ending A-roll; never match omitted candidates; bridge only the selected brief A-roll flashes; mask only picture-reviewed objectionable jump cuts; never reuse an asset SHA-256 or visual identity; run filename matching for every selected B-roll need and pass its match receipt into source-window selection; validate and describe the selected manifest-backed source window; when BGM is requested, select timestamps from pi-media waveform PNGs, default to 12 LU below dialogue before sidechain ducking, and derive gain from exact A-roll/BGM EBU R128 manifests instead of a percentage; and export a generic pi-media EDL for this request: $@

Show the proposed rhythm before writing a new revision. Never overwrite the source or an existing render.