What problem does it solve? Editing talking-head videos manually is slow: you must scrub footage to find filler words, stutters, retakes, and silences, cut them without losing meaning, then rebuild subtitles and assemble the final video. This Skill automates that two-stage workflow from raw recording to exported MP4. ## Core Features & Use Cases - Speech Cleanup (Stage 1): Transcribes the raw video with word-level timestamps, detects deletion candidates (silences, filler words, stutters, repeated phrases, broken sentences, retakes), generates an HTML review page, and cuts a clean video with ffmpeg after user confirmation. - Risk-Tiered Deletion Rules: Applies a "delete earlier, keep later" principle and risk-tiered review so low-risk cuts (silences, short fillers) go straight to review while high-risk cuts (whole-sentence deletions) get adversarial verification. - Final Assembly (Stage 2): Builds a storyboard from the cut video, subtitles, and assets, previews a second-by-second timeline in HTML, and exports the final MP4 at the configured aspect ratio (e.g., 3:4 vertical). - Use Case: You recorded a 20-minute talking-head video full of "um", stutters, and repeated takes. The Skill transcribes it, proposes cuts, lets you confirm them in a review page, outputs a clean source_cut.mp4 with AI-proofread subtitles.srt, then assembles the final vertical video. ## Quick Start Ask the AI to cut this talking-head video by removing filler words and mistakes, then export the final vertical MP4 with subtitles.
