What problem does it solve? Finished videos (HyperFrames explainers, slide decks, screencasts) often have narration that drifts from the picture: the voiceover talks about the wrong chapter, runs shorter or longer than the visual segment, or crosses chapter title cards. This Skill repairs the audio and subtitle track of an already-rendered video without re-rendering the visuals. ## Core Features & Use Cases - Misalignment Diagnosis: Runs a 1fps frame probe to build a visual chapter inventory, then classifies the problem as topic drift, length drift, or boundary drift before rewriting anything. - Script Rewrite + TTS Alignment: Rewrites the voiceover script per visual chapter, then uses ffmpeg atempo compression or silence padding to fit each chapter's TTS audio to its target duration. - Subtitle Generation Without Whisper: Extracts sentence-level timings via ffmpeg silencedetect, reconciles phrase counts against the script, and burns SRT subtitles into the MP4 with the ffmpeg subtitles filter. - Use Case: A 14-minute Chinese course video has a TTS track 30 seconds short and subtitles showing "Day 05" while the screen shows "Day 04". This Skill probes the frames, rewrites the script to match the visuals, aligns each chapter's audio, and produces a final MP4 with accurate burned-in subtitles. ## Quick Start Ask the AI to fix the audio and subtitles of your rendered video so the narration matches what is on screen, for example: "My course video's voiceover is out of sync with the chapters — realign the narration and burn in corrected subtitles."
