How to Give Claude Eyes to Watch Any Video Frame by Frame (Free & Local)

👁️ Give Claude Eyes: Watch & Analyze Any Video, Frame by Frame

Turn Claude from a transcript-reader into a full-on video analyst. No more missing half the details — this setup lets Claude actually see what's happening on screen.

🎯 Overview

By default, Claude can only read a video's transcript. That means it misses everything visual — on-screen text, gestures, product shots, UI walkthroughs, scene changes, all of it. You lose roughly half the story.

With two lightweight skills, you flip that on its head:

1. Grab the video — pull down any reel, short, or long-form clip.
2. Rip the frames — split it into individual images Claude can actually look at.

Then Claude breaks the whole thing down frame by frame.

💡 It runs 100% locally, so there are zero API costs and it's insanely fast — even on a 30-minute video.

⚡ What You Can Do With It

🛠️ The Two Skills

Both run on your machine — no uploads, no per-call fees, no waiting on external servers.

🚀 Step-by-Step Setup

1️⃣ Grab the video

Point the first skill at your video link. It pulls the full file down locally.

2️⃣ Rip the frames

Run the second skill on the downloaded file. It slices the video into a folder of image frames.

💡 Tune your frame interval based on the video. Fast-paced reels? Grab more frames per second. Slow talking-head content? Fewer frames keeps things lean and fast.

3️⃣ Feed the frames to Claude

Hand Claude the frames (plus the transcript if you've got it) and ask for a breakdown.

📝 Prompt Library

Copy, paste, and tweak these once Claude has the frames.

🔍 Full Frame-by-Frame Breakdown

You now have both the transcript and a sequence of extracted frames from this video.

Go through it chronologically and give me a frame-by-frame breakdown. For each key moment include:
- Timestamp (approx)
- What's happening visually (on-screen text, actions, scene)
- What's being said (from the transcript)
- Why this moment matters to the video's goal

Flag anything visual that the transcript alone would have missed.

🪝 Hook Teardown (Reels & Shorts)

Analyze the first 3 seconds using the frames provided.

Break down:
- The exact visual hook (what fills the screen)
- On-screen text vs. spoken words
- Pattern interrupt or scroll-stopper being used
- Why it works (or doesn't)

Then suggest 3 stronger hook variations based on what I could shoot instead.

🧩 Tutorial Extraction

This is a tutorial video. Using the frames, extract the exact step-by-step process shown on screen.

For each step:
- Number it
- Describe the on-screen action / UI state
- Note any settings, values, or clicks visible in the frame

Output as a clean numbered checklist I could follow without watching the video.

📊 Ad / Product Breakdown

Using the frames and transcript, deconstruct this as a marketing asset.

Cover:
- Product shots and how the product is framed
- Text overlays and their timing
- The persuasion structure (problem → solution → proof → CTA)
- Visual transitions and pacing

End with what's working and one thing I'd change.

Based on the full video (frames + transcript), write me an original short-form script inspired by its structure — not a copy.

Keep the same pacing and hook style, but rewrite it in my voice for my topic: [YOUR TOPIC].

Deliver as: Hook, Body, Payoff. Under 45 seconds spoken.

📈 Before vs. After

💎 Pro Tips

💡 Keep the transcript in the mix. Frames + transcript together give Claude both the what's said and the what's shown — that combo is where the magic is.

✅ Quick Recap

1. 📥 Grab the video (local)
2. 🖼️ Rip it into frames (local)
3. 👁️ Feed frames + transcript to Claude
4. 📝 Run a prompt and get a full breakdown

That's it — Claude can finally see the whole video.