skills.rest
three1324
three1324COMMUNITY
Explore all skills by this author →
GitHub

watch

Extract frames and transcripts from videos using yt-dlp, ffmpeg, and Whisper APIs.

Updated Aug 16, 2026
Search Tags:#youtube#transcription#yt-dlp#whisper#ffmpeg#video analysis#frame extraction
One-click install
npx skills add https://github.com/three1324/yeonjinautomotive --skill watch-three1324
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: watch
Source: https://github.com/three1324/yeonjinautomotive/tree/main/.claude/skills/watch
Command: npx skills add https://github.com/three1324/yeonjinautomotive --skill watch-three1324

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires yt-dlp, ffmpeg, and includes scripts (resource) components.

What problem does it solve? AI assistants cannot natively view video content, so they cannot answer questions about a YouTube link or a local video file. This Skill downloads the video, extracts timestamped frames, and pulls a transcript so the AI can see and hear what happens in the video. ## Core Features & Use Cases - Frame Extraction: Scene-aware, keyframe, or uniform frame sampling via ffmpeg with four detail modes (transcript, efficient, balanced, token-burner) and a 2 fps cap to control token cost. - Transcript Retrieval: Pulls native captions through yt-dlp first, falling back to Groq or OpenAI Whisper API transcription when captions are missing. - Focused Analysis: Supports --start/--end time ranges and --timestamps for grabbing frames at transcript-flagged moments like "look here" cues. - Use Case: Paste a YouTube URL and ask "what does the presenter show at 2:30?" — the Skill downloads the video, extracts dense frames around that moment, filters the transcript to the range, and lets the AI answer with visual evidence. ## Quick Start Ask the AI to watch a video by providing a URL or local file path along with your question, for example: watch https://youtu.be/abc123 and summarize what happens.

Frequently Asked Questions about watch

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I let Claude watch and analyze a YouTube video?▼

Provide the video URL with your question, and the watch script downloads it via yt-dlp, extracts scene-aware frames with ffmpeg, and pulls the transcript from captions or Whisper. Claude then reads the frame images and transcript to answer.

What tools are needed to extract frames from video files?▼

This Skill requires ffmpeg and ffprobe for frame extraction and yt-dlp for downloading videos and captions. On macOS the setup script installs them automatically via Homebrew; on Linux and Windows it prints the exact install commands.

Does video transcription work without a Whisper API key?▼

Yes, when the source platform provides native captions, yt-dlp pulls them for free without any API key. Without captions and without a Groq or OpenAI key, the Skill returns frames only and notes the limitation.

How do I analyze a specific section of a long video?▼

Pass --start and --end timestamps in SS, MM:SS, or HH:MM:SS format to focus extraction on that range. Focused mode samples frames more densely, up to 2 fps, and filters the transcript to the same window.

What are the limitations of AI video frame analysis?▼

Accuracy is best for videos under 10 minutes since frame coverage thins with duration, and sampling never exceeds 2 fps. Frame count is capped by detail mode (50 for efficient, 100 for balanced), so fast motion between frames can be missed.

Related Skills

Explore More Content & Communication Skills→
azmiariffarisazmiariffaris

id-social-media-scan

Scan Yahoo search results for public Hermes Agent content on Indonesian platforms.

Community
Intermediate
AllBeingsFutureAllBeingsFuture
9

allbeingsfuture/skills@youtube-summarizer

Extract YouTube transcripts and generate timestamped summaries with STAR and R-I-S-E analyses.

Community
Intermediate
patrykk21patrykk21

fact-check

Extracts YouTube transcripts, identifies key claims, and validates them against peer-reviewed sources.

Community
Intermediate
gyalamanch001agyalamanch001a

video-understand

Extract frames and transcribe speech from video files into structured JSON.

Community
Advanced