claude-video
Give Claude eyes and ears: paste a video link, and it extracts frames, transcribes audio, then hands all that context to Claude—helping developers and content creators quickly understand tutorials, recordings, and demos.
Tool overview
Adoption verdict: The project blew up on GitHub Trending in July 2026, gaining 368 stars on its first day and 3.8k stars in total. Multiple sources include actual hands‑on reports—one Zhihu article listed detailed parameters and said the experience was “more complete than expected”. The strong social media hype is backed by some first‑hand testing; it’s an innovative early‑stage tool worth watching but not yet production‑ready. How it works: Use the /watch command with a video URL (e.g. YouTube). The pipeline: download → ffmpeg scene‑aware frame extraction (tunable via --scene, --fps-floor, --max-frames, etc.) → Whisper transcription with timestamps → merge frame descriptions and text into one long document → submit to Claude. You can ask questions like “what happens at 1:30?” and Claude answers by correlating the relevant frame and audio snippet, effectively turning the video into an analysable text corpus. Threshold and cost: Requires local installation of ffmpeg, Python and dependencies; some CLI skill is needed. You must provide your own Claude API key; costs depend on video length and frame density.