skills.rest
mintuz
mintuzCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

watch-video

Transcribe and analyze videos from YouTube, Loom, Vimeo, and local files at selectable depth.

3|Updated Nov 8, 2014
Search Tags:#youtube#yt-dlp#whisper#ffmpeg#loom#video transcription#multimodal analysis
One-click install
npx skills add https://github.com/mintuz/.dotfiles --skill watch-video-mintuz
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: watch-video
Source: https://github.com/mintuz/.dotfiles/tree/main/agents/.agents/skills/watch-video
Command: npx skills add https://github.com/mintuz/.dotfiles --skill watch-video-mintuz

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires yt-dlp, ffmpeg, mlx-whisper.

What problem does it solve? Watching long videos to extract key information is time-consuming, and different videos need different levels of analysis. This Skill automates transcription and analysis of videos from YouTube, Loom, Vimeo, Riverside, Zoom recordings, social platforms, and local files, letting you choose between a fast free transcript, a visual frame-by-frame analysis, or full multimodal AI ingestion. ## Core Features & Use Cases - Three depth modes: transcript (fast, free text extraction), visual (transcript plus ffmpeg frame extraction and Claude vision pass on key moments), and multimodal (Gemini native video ingestion or dense Claude vision). - Multi-source support: Downloads via yt-dlp from YouTube, Loom, Vimeo, X/IG/TikTok, or accepts local MP4/MOV/WebM files, preferring platform-provided transcripts before falling back to local MLX-Whisper transcription. - Structured outputs: Saves transcript, metadata, key moments, and a summary with flagged action items and decisions to an organized workdir, with optional capture to a second-brain notes vault. - Use Case: A teammate shares a 45-minute Loom walkthrough of a new feature. Run the visual mode to get a transcript, timestamped key moments of UI changes, and a summary with action items — without watching the full recording. ## Quick Start Ask the AI to watch and summarize a video by providing its URL, for example: transcribe and summarize this Loom recording at https://www.loom.com/share/abc123 using visual mode.

Frequently Asked Questions about watch-video

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

FAQPage Schema
How do I transcribe a YouTube video without watching it?▼

Provide the YouTube URL and the default transcript mode downloads auto-generated subtitles via yt-dlp, cleans the rolling captions, and saves a plain-text transcript. If no subtitles exist, it falls back to local MLX-Whisper transcription.

How to extract key moments from a Loom recording?▼

Use visual mode on the Loom URL. It downloads the video at 720p, extracts frames with ffmpeg at a cadence suited to screen-shares (about one frame per 5 seconds), and runs a Claude vision pass to identify UI changes and notable moments with timestamps.

What is the difference between transcript, visual, and multimodal modes?▼

Transcript mode returns only cleaned text and metadata, fast and free. Visual mode adds ffmpeg frame extraction plus a Claude vision pass on key moments. Multimodal mode sends the video natively to Gemini if an API key is set, otherwise uses dense Claude vision frame analysis.

Does video transcription work without a Whisper API key?▼

Yes. The skill uses MLX-Whisper running locally on Apple Silicon Macs, so no cloud API key is needed. It also prefers free platform-provided transcripts from YouTube, Loom, and Riverside before invoking Whisper at all.

Why does multimodal mode warn about cost on long videos?▼

Multimodal analysis ingests the entire video through Gemini or dense per-frame Claude vision, which is meaningfully expensive on videos over 10 minutes. The skill warns before running and offers transcript-only or visual mode as cheaper fallbacks.

What tools are required to analyze local MP4 files?▼

You need yt-dlp for URL downloads, ffmpeg for frame extraction in visual and multimodal modes, and mlx-whisper for transcription when no platform transcript exists. Local files are probed with ffprobe and processed directly without downloading.

Related Skills

Explore More Content & Communication Skills→
imMamdouhaboammarimMamdouhaboammar

watch-video

Transcribe and analyze videos from YouTube, Loom, Vimeo, and local files at selectable depth modes.

Community
Advanced
asonasasonas
5

wrapup

Summarize daily work from Things3, conversation, and second-brain into an Obsidian daily note.

Community
Advanced
sebaperiesdigit-createsebaperiesdigit-create

record-a-skill-custom

Converts demonstrated workflows and evidence into confirmed reusable skill specifications.

Community
Advanced
OpenLaborOpenLabor
4

Meeting Action Extractor

Parse meeting transcripts to extract decisions, actions, owners, deadlines, questions, and follow-ups into structured data.

Official
Advanced