Record on a phone, drop the file in a folder, get back a finished reel. Claude cuts every silence and dead-air pause, transcribes the audio word by word, and burns TikTok-style captions synced to each spoken word, rendered with Remotion, no script ever typed.
“Process the video in the Raw folder” is the whole instruction. Claude finds the clips, stages them to the cloud workspace, and rebuilds the toolchain (whisper + Remotion) if the session is fresh. For multi-take projects it works audio-first, transcribing every take and keeping the last attempt at each line before moving any heavy footage.
Whisper transcribes the audio with word-level timestamps. ffmpeg detects every silence and dead-air pause, cuts them, and re-stitches the keeper segments. The word timings are remapped onto the cut timeline so Remotion can burn bold, uppercase captions with the current word highlighted, the style the algorithms read.
The draft MP4 and transcript come back for a caption skim, the only human step. Mis-heard words are fixed in the caption data and only the render re-runs. On approval, the full-quality final lands in a Completed project folder with the transcript and original takes, and Raw is left empty for the next reel.