How to Make AI Watch a YouTube Video
Co-founder at King’s Cross Labs · ex-LinkedIn PM & Forbes 30 Under 30
Send this guide to yourself
Get the link in your inbox so you can read it whenever you're ready.
Your email will also be saved for future updates.
This is a skill that lets Claude or Codex watch long videos for me. I paste a YouTube link instead of spending the entire hour watching the full thing.
If you're a creator, marketer, founder, or just someone who wants to learn better, this is for you. I use it when I want to analyze other creators in the space, or learn something new for my business or my life.
Need Claude Code first? Here's the 5-minute setup. New to skills? Start with What is a Skill?.
How do you download it from GitHub?
The skill lives here: github.com/jordanrendric/claude-video-vision
Skip the zip download. Point Claude Code at that GitHub repo and install it as a plugin, the same way you'd install a marketing skill pack.
1. Open the GitHub repo.
Go to github.com/jordanrendric/claude-video-vision and copy the URL. That's the source you'll add in Claude Code.
2. Open Claude Code.
If you don't have it yet, set up Claude Code first. Regular claude.ai chat cannot install this plugin.
3. Add the GitHub repo as a plugin marketplace.
In Claude Code, run this by itself, then wait for it to finish:
/plugin marketplace add https://github.com/jordanrendric/claude-video-visionThis tells Claude Code it can install plugins from that GitHub repo. Don't paste the next command in the same message.
4. Install the plugin.
/plugin install claude-video-visionRun that by itself too. The first time you use it, the video server installs in the background via npm. You don't build anything.
5. Run the setup wizard.
/setup-video-visionIt checks for ffmpeg (pulls the frames) and yt-dlp (lets you paste a YouTube URL instead of a file on your computer). If either is missing, the wizard tells you what to install. On a Mac that's usually brew install ffmpeg and brew install yt-dlp.
It also asks which audio backend you want. Gemini is the easiest free start. You need an API key from ai.google.dev. Local Whisper stays on your machine. OpenAI Whisper is paid.
I also run /watch-video in Codex once the skill is on that machine. The two /plugin commands above are the Claude Code path, which is how this GitHub repo is meant to be installed.
If you already clone repos: git clone that URL, then start Claude Code with --plugin-dir pointed at the folder. Most people should use the /plugin commands instead.
What do you type after it's installed?
/watch-video [youtube URL]One command, one link. A question after the URL gets a better answer:
/watch-video https://www.youtube.com/watch?v=... "tell me why the opening works"You can pass a local file too, like ~/Downloads/tutorial.mp4, if you already have the video.
It analyzes frame by frame what's happening. That's how it can tell me which spoken and visual hooks work, why the editing works, or turn the video into a playbook so I learn faster.
If you want a study loop from a playlist (outline, quiz, tutor), that's How to Turn a YouTube Playlist into an AI Tutor. That one is sources and practice. This one watches, including what was on screen.
Key insight: It looks at frames and audio together, so it can talk about the visual hook and the cut.
What can you use watch-video for?
Three jobs I use it for, with the prompt I'd paste.
Analyze other creators
This is the one I reach for first. Paste a competitor or a creator you respect. Ask for the parts a caption file never mentions: what's on screen in the first three seconds, where they cut, what the text overlay says.
/watch-video [YouTube URL] "Break down the spoken hooks and the visual hooks. What is on screen in the first 3 seconds? Why does the editing work? What would I steal for my own content?"Learn something without sitting through the hour
Long tutorial, lecture, product demo, conference talk. Tell it what you need to walk away with. Ask for the framework and what to try this week, not a recap of the whole video.
/watch-video [YouTube URL] "I don't need a recap. Give me the framework, the steps in order, and what I should try this week."Turn a video into a playbook
Once it has watched, ask it to package the useful parts into something you can run again, instead of a summary you won't reopen.
/watch-video [YouTube URL] "Turn what you watched into a playbook I can run: steps, the prompts I'd reuse, and what good looks like at each stage."Here are some related guides to check out:
Frequently asked questions
- Can I do this if I'm not technical?
- Yes. You need Claude Code or Codex, then two plugin commands from the GitHub repo and a setup wizard that checks ffmpeg and yt-dlp for you. You paste a YouTube link and ask a real question. You do not write any code.
- How is this different from pasting a YouTube transcript into Claude?
- A transcript is the spoken words. This skill extracts frames and audio, so Claude can see the on-screen text, the jump cuts, and the first three seconds. Use /watch-video when the visual is part of why the video works.
- Does /watch-video work with TikTok or Instagram, or only YouTube?
- YouTube URLs work as a paste-a-link path. Local video files work too, so you can download a TikTok or Reel as an mp4 and pass the file path. I would not count on pasting an Instagram or TikTok URL the same way you paste a YouTube link.
- Do I need Claude Code, or does this work in regular Claude chat?
- You need an agent that can install the plugin and run the video tools, which for me is Claude Code and Codex. Regular claude.ai chat cannot install this plugin or extract frames off a YouTube URL on its own. Set up Claude Code first, then come back to this.
Want to build your first AI agent?
Build Your First Agent 101 is the next step! You'll get step by step guides, video tutorials and starter prompts to create your first agent in half a day, customized to your own workflow.
Learn more