Skip to content

Claude-real-video － any LLM can watch a video

8.1 relevance
Score Breakdown
technical depth
8
novelty
9
actionability
8
community
7
strategic
6
personal
10

Scored daily by a customisable AI persona to surface the most relevant engineering leadership news.

Claude-real-video enables any LLM to watch video, highly relevant to AI/ML and open source.

AI/ML github.com
Let Claude (or any LLM) actually watch a video — scene-aware, deduplicated frames + transcript, from a URL or local file. Runs locally, MIT. - HUANGCHIHHUNGLeo/claude-real-video
Summary

Open-source claude-real-video uses scene-change detection (sensitivity 0.30) and sliding-window dedup (threshold 8%) to extract only meaningful frames from videos, then combines them with Whisper transcriptions and a manifest for any LLM — no cloud upload, no fixed-interval sampling. It handles URLs (via yt-dlp) or local files, stays local, and defaults to a 150-frame cap, giving agents richer, cheaper context for video understanding.

Author

HUANGCHIHHUNGLeo