FREE · SETUP GUIDE

Give Claude Eyes and Ears.

Claude can't watch or hear videos; paste a reel link and it grabs the caption but misses everything that made it go viral. Three free tools fix that, locally, in about 20 seconds per reel.

Phone is optional. By adding it you agree Ash Harris Productions may call or text you about this; message and data rates may apply, and you can opt out any time.

You'll also receive my content strategy newsletter with more free breakdowns.

Step-by-Step Guide

The 3-Tool Stack That Lets Claude Watch Reels

Open Claude Code and install the stack with one prompt

Three tools, one install. yt-dlp downloads the video, Whisper transcribes the audio, and FFmpeg slices it into frames Claude can actually look at. Claude Code sets all of it up for you; paste the prompt below into a fresh session in an empty folder. Everything runs locally, so there are zero API costs.

Install yt-dlp, ffmpeg, and whisper.cpp on this machine so you can download, transcribe, and frame-analyse videos for me. Verify each one works before finishing.

Get Claude Code→

Paste the reel link; yt-dlp grabs the video

Drop any Instagram, TikTok, or YouTube link into your prompt. yt-dlp pulls the full video file plus the caption and metadata, so Claude starts from the real thing instead of a caption summary.

Use yt-dlp to download this reel into ./reference and keep its metadata: <paste the reel link>

Whisper gives Claude ears

Whisper transcribes the audio locally with word-level timestamps. Claude now hears the delivery: the exact hook wording, where the pauses land, and how fast the creator talks in the first three seconds. This is the part the caption never shows you.

FFmpeg gives Claude eyes

FFmpeg exports a frame every second. Claude reads them alongside the transcript, so it catches the on-screen text, the graphics, the cuts, and the b-roll changes, and it understands how all the sections work together.

Extract one frame per second with ffmpeg, then read the frames alongside the transcript. Give me the hook, the pacing map, and every graphic on screen with timestamps.

Get the 20-second breakdown, then your version

This is the payoff. I fed it a reel with 100k views and 20 seconds later it handed me the hook, the pacing, and every graphic on screen. From there, one more prompt turns the breakdown into your script: same structure, your story, your offer.

Save the breakdown to a file each time; a folder of these becomes a reference library Claude can pull patterns from.

Let Claude edit its own version with HyperFrames

Analysis is half the loop. With HyperFrames installed, Claude takes the script it just wrote and cuts the video too: captions, pacing, graphics, rendered to a finished 9:16 reel. Pasting a reel link in and getting an edited draft out is now a single Claude Code session.

The HyperFrames setup guide→

Enter your email to unlock