#3737th of 69 in Data and scraping for AI

claude-real-video

Extracts scene-change keyframes and a timestamped transcript from videos for LLMs

Stars
2.3k
License
MIT
Last commit
Oct 2026
Last release
Sep 2026
Language
Python
Stars, last 19 weeks+1,534
4 Jul 2026: 790 stars11 Jul 2026: 1,521 stars18 Jul 2026: 1,726 stars25 Jul 2026: 1,851 stars1 Aug 2026: 1,907 stars8 Aug 2026: 1,972 stars15 Aug 2026: 2,017 stars22 Aug 2026: 2,049 stars29 Aug 2026: 2,085 stars5 Sept 2026: 2,119 stars12 Sept 2026: 2,133 stars19 Sept 2026: 2,161 stars26 Sept 2026: 2,186 stars3 Oct 2026: 2,200 stars10 Oct 2026: 2,214 stars17 Oct 2026: 2,242 stars24 Oct 2026: 2,265 stars31 Oct 2026: 2,280 stars7 Nov 2026: 2,300 stars14 Nov 2026: 2,324 stars

790 on 4 Jul 20262,324 on 14 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

claude-real-video (crv) takes a URL or local file, keeps frames at scene changes with sliding-window dedup, transcribes audio with Whisper, and writes frames, frames.json, transcript and MANIFEST.txt that any LLM can read. It runs locally through a CLI, a local web page, a Python API, an agent skill and an MCP server, and indexes past analyses in a local SQLite file for search. The free edition omits on-screen text recognition and shot analysis, which are sold in paid add-ons.

Who it is for: Users of Claude Code or other agents who want video turned into frames and text

Strengths

  • Scene-change selection with dedup: a 58 s clip gives 26 frames instead of 58 at 1 fps
  • Processing is local; the source video is not uploaded anywhere
  • Several Whisper backends: openai-whisper, faster-whisper with VAD, and mlx-whisper on Apple Silicon
  • Ships as CLI, Python API, agent skill and MCP server with five tools

Weaknesses

  • On-screen text recognition and shot or emotion analysis are paid add-ons, not in this edition
  • Requires a separately installed ffmpeg; no speech-to-text without the whisper extra
  • Frames are pasted into an LLM afterwards; vision understanding depends on that model
  • Default 640 px frames lose small text unless --frame-width is raised

What it needs

  • GPU optional
  • Needs ffmpeg, ffprobe, yt-dlp, whisper
  • Models: openai-whisper, faster-whisper, mlx-whisper, Silero VAD, speaker diarization (45 MB)

What changed

From the release notes. No flag means the note does not say so, not that the upgrade is safe.

Also in Data and scraping for AI

See all 69
Also in Data and scraping for AI
RankProjectScore
1mineruParses PDFs, images and Office files into Markdown and structured JSON Live demo ↗ (opens in a new tab)83.9k starscustom license91 out of 100
2doclingPython library that converts PDFs, Office files and more into structured text70.7k starsMIT91 out of 100
3opendataloader-pdfPDF parser that outputs Markdown, JSON with bounding boxes, and Tagged PDF30.3k starsApache-2.091 out of 100
#4agent-reachCLI that gives coding agents read access to login-gated and Chinese web platforms112.8k starsMIT88 out of 100
#5markitdownPython library and CLI that converts office files, PDFs and audio to Markdown196.2k starsMIT87 out of 100