claude-real-video-for-agents
Claude Real Video lets AI agents process videos by intelligently extracting scene-aware keyframes and audio transcripts. It uses scene-change detection and deduplication to reduce token usage—a 58-second clip yields 26 meaningful frames instead of 58, packed into contact sheets for efficient LLM consumption.
Claude Real Video for Agents extracts and deduplicates keyframes from videos so AI agents can analyze them.
AI-generated summary based on this skill's SKILL.md
Decision gist · record as of 2026-07-27
Claude Real Video for Agents extracts and deduplicates keyframes from videos so AI agents can analyze them. Claude Real Video lets AI agents process videos by intelligently extracting scene-aware keyframes and audio transcripts. It uses scene-change detection and deduplication to reduce token usage—a 58-second clip yields 26 meaningful frames instead of 58, packed into contact sheets for efficient LLM consumption.
Use it when
- Claude Real Video uses scene-change detection to identify visually distinct moments in your video.
- Yes.
Verify before relying
Read SKILL.md below before installing (1 file). Open directory: indexed for reading, not audited.
Install
HUANGCHIHHUNGLeo/claude-real-video/claude-real-video-for-agents · repository language: Python
Open directory. Skills are indexed for reading, not audited. Review a skill's body before installing it.
Frequently asked questions
AI-generated answers based on this skill's SKILL.md and metadata
How do I make AI agents watch videos with claude-real-video?
Claude Real Video enables agents to process videos by extracting scene-aware keyframes and audio transcripts. The tool uses scene-change detection to identify meaningful moments, then deduplicates frames and packs them into contact sheets. This approach lets agents understand video content without processing every single frame, reducing token usage significantly.
How does claude-real-video extract keyframes from video?
Claude Real Video uses scene-change detection to identify visually distinct moments in your video, then extracts keyframes at those transition points. It automatically deduplicates similar frames so you get only the meaningful ones. For example, a 58-second clip typically yields 26 keyframes instead of 58, capturing scene changes rather than redundant content.
Can I convert video to frames and transcript for multimodal analysis?
Yes. Claude Real Video transcribes audio using Whisper and pairs the transcript with extracted keyframes. This multimodal pairing lets agents understand both visual and audio content together, enabling richer analysis of video material for your AI workflows.
Does claude-real-video process videos locally without uploading?
Claude Real Video processes local or URL-based videos offline without uploading to external services. You maintain full control over your video data while the tool handles keyframe extraction, deduplication, and transcription on your machine.
What is crv installation and how do I set it up?
Claude Real Video can be installed as an agent skill (crv) to enable video analysis workflows directly in your agent pipelines. Installation integrates the tool into your agent environment, allowing automated video processing tasks like frame extraction, transcription, and contact sheet generation as part of your agent's capabilities.
How does claude-real-video reduce token usage for LLM consumption?
Claude Real Video packs deduplicated keyframes into contact sheets—grid layouts that combine multiple frames into single images. This dramatically reduces the number of tokens needed compared to processing individual frames. Scene-aware deduplication ensures you send only meaningful visual content to your LLM, maximizing efficiency.
SKILL.md
Rendered from the published skill. Quoted content, verbatim.
claude-real-video for AI agents
What is crv?
crv (claude-real-video) is a CLI tool that extracts meaningful frames and transcripts from videos so AI agents can "see" and "read" them. It uses scene-change detection (not fixed-interval sampling), sliding-window deduplication, and optional Whisper transcription.
Key advantage: Same 58-second clip at fixed 1fps = 58 frames. crv keeps the 26 that actually differ, and --grid packs them into 3 contact sheets. Fewer tokens, nothing missed.
Installation
Prerequisites
- Python 3.10+
- ffmpeg / ffprobe on PATH
# macOS
brew install ffmpeg
# Linux
sudo apt install ffmpeg
# Windows
winget install Gyan.FFmpeg
Install crv
```bash
Recommended: with audio transcription support
pip install "claude-real-video[whisper]"
Core only (frames + dedup)
pip install
(truncated - see the full file via the links below)
File tree — 1 file
skills/claude-real-video-for-agents/SKILL.md
Let your AI agent find skills like this
Example. Real query, live index.
You found this page by searching. An agent finds it by wishing: SkillFed indexes 56,283 agent skills by what they can do, searchable in plain language.
wish › “Extract and deduplicate keyframes from videos so AI agents can analyze visual content”
Give your agent the search over MCP, or paste the wish link into any chat. No install? Search from any chat →
Related skills
claude-real-video bridges Claude's text-only interface with video content by automatically extracting scene-aware keyframes and speech-to-text transcripts. Pass a video URL or file path, and the skill generates contact sheets and a transcript that Claude can then read to answer your questions, summarize the content, or perform detailed analysis. Local processing keeps your source video private.
watch lets you analyze any video by intelligently sampling frames at scene changes, measuring editorial pacing (cuts per minute, shot length), and applying dense analysis to the critical first 10 seconds. It pulls transcripts from captions or Whisper, generates a structured report, and optionally saves findings to your Obsidian vault.
Watch Skill processes videos locally by extracting scene-aware frames, running OCR on text within them, and transcribing audio via offline Whisper or captions. It builds a searchable index so follow-up questions are answered without re-processing the video. No API keys required for core features; vision and STT are optional and configurable.
Watch lets you analyze any video—from YouTube, local files, or other sources—by automatically pulling frames and transcripts so Claude can answer your questions about the content. It handles caption extraction, falls back to Whisper when needed, and scales frame detail to match your needs.
Watch Skill processes any video—URLs, local files, streams, recordings—by extracting frames, running OCR, and transcribing audio into a unified searchable index. Query the index directly instead of re-watching, and batch-process playlists or folders in one pass.
Fetch videos and audio directly from URLs using yt-dlp, supporting YouTube and over 1000 other platforms. Extract audio tracks, retrieve captions in multiple languages, or inspect metadata without storing files. Customize resolution, format, and output paths to fit your needs.