Source: bradautomates/claude-video | MIT | 18.6K stars | last push 2026-09-25.
What it does. Downloads a video (URL or local file) with yt-dlp, pulls frames with ffmpeg, grabs the transcript from captions (falling back to a Whisper API if you give it a key) and hands frames plus transcript to Claude so you can ask questions about the video.
What it needs. Python, ffmpeg, ffprobe and yt-dlp. The Whisper fallback is optional and uses your own API key.
What we checked. On a real YouTube video it downloaded, extracted 19 frames and returned the real caption transcript, and we opened a frame to confirm the content matched. We did not test long videos or videos without captions.
Who it suits. Research, summarising talks, checking what a competitor video actually shows. Watch out: one setting uses an ffmpeg flag removed in ffmpeg 9, so it works on 8.x and needs a small edit after a major upgrade. Downloading videos may be against a site’s terms; check before you do it at scale.
Get the full skills list
By email, including rejects.
Sascribe is independent. Figures come from the linked primary pages on the date shown and can change. Nothing here is financial advice or a promise of income.