Description
Clean single-speaker and turn-taking spoken videos while preserving their meaning. The AI host reviews timed transcripts, marks complete phrases to keep or remove, and controls either ChatCut Desktop or a local FFmpeg pipeline. Outputs include an edited MP4, synchronized SRT when enabled, an edit plan, timeline data and machine-check reports. Repeated runs reuse matching output and stop on conflicting manual changes. The local route requires Python 3.10+, FFmpeg/ffprobe and libx264; subtitle burn-in also requires libass. ChatCut requires a separately installed, authenticated Desktop app and its official local connection. Timed transcripts may be supplied directly; optional local faster-whisper needs a separately supplied model. The local route rejects overlapping speech and does not provide multi-camera editing or complex rearrangement. Six local behavior tests, including actual rendering and subtitle burn-in, passed in this run. ChatCut, optional ASR, real-person lip sync, long videos and human full playback were not tested in this publishing run.