
Description
video-understand is an agent skill from the OpenMontage project that describes what is in a video, fully offline and without any API keys. It extracts frames with ffmpeg and transcribes the speech with Whisper, so the assistant can work out what a video shows and what is said in it. The default command runs scene detection and transcription, and a keyframe mode extracts frames at key points instead. The skill needs ffmpeg and ffprobe, which are required, and the openai-whisper Python package for transcription, which is optional. You install ffmpeg with Homebrew and whisper with pip, then run the script on your video file. It is useful for reviewing a recording, summarizing a talk, finding a moment by what is said, or checking what a clip contains before you edit it. Because transcription runs locally, long videos take time, so start with a short clip to check the setup. The skill passes Gen Agent Trust Hub, Socket and Snyk on skills.sh, and it has about 1,200 installs. Install it with npx skills add https://github.com/calesthio/openmontage --skill video-understand.…
Highlights
- Extracts frames and transcribes speech fully offline
- Scene detection and transcript by default
- Keyframe extraction mode
- No API keys required
Ratings & Reviews
No ratings yet
Be the first to rate video-understand.
People also visit
video-understand FAQ
What does it need?
ffmpeg and ffprobe are required. The openai-whisper package is optional and is needed for transcription.
Does it need an API key?
No. The skill runs fully offline.
What modes are there?
Scene detection with transcription by default, and keyframe extraction with the keyframe mode.
How long does it take?
Transcription runs locally, so long videos take time. Start with a short clip to check the setup.
How do I install it?
Run npx skills add https://github.com/calesthio/openmontage --skill video-understand.
What are the best video-understand alternatives?
Popular alternatives to video-understand include openai-whisper, video-edit and hyperframes-media. See all video-understand alternatives







