Study a reel. Learn what works. Make your next video your own.
A reel link alone may not give Claude the video to study. This workflow brings it the frames, transcript, and context—so you can learn from the hook, pacing, visuals, and CTA.
Claude● ● ●
YOU
Break this down for me.
instagram.com/reel/example
CLAUDEBLOCKED
I can’t access that link.
Illustrative example · access varies by tool and post
⌕
THE WORKAROUND
Give it the reel’s actual content.
↗
FROM A LINK TO A LESSON
Four steps. One useful workflow.
The tools prepare the material. You and Claude turn it into lessons for original content.
1
Download
Instaloader / yt-dlp
Your permitted public reel
2
Keyframes
ffmpeg
Frames + a contact sheet
3
Transcript
Whisper, locally
Words with timestamps
4
Claude Code
Analyze + learn
Principles for your next reel
WHAT’S INSIDE
A setup you can actually use.
The commands, the prompts, and a repeatable way to learn from your saved reels.
2 free downloadersLocal transcription4 prompts
The install commands, with official references
The prompt that builds your local workflow
The hook, pacing, visuals, and CTA breakdown
The “make it mine” prompt for original work
A saved-links workflow for comparing patterns
KEEP THE SETUP HANDY
Ready to look closer?
Start with one reel you have permission to study.
THE STEP-BY-STEP SETUP
Build your Reel Lab.
Mac-first instructions. Commands checked against official project documentation on November 17, 2026.
↳
Your guide is one step away.
Use either signup form to unlock the complete setup and copyable prompts.
A Claude subscription or account access that includes Claude Code; usage follows your plan.
Homebrew and Python 3.11+ for current yt-dlp. The commands below use Python 3.12 in a dedicated environment.
Disk space for videos, frames, and Whisper model downloads. Setup and transcription time depend on your computer.
Windows? Use Anthropic’s Windows/WSL instructions ↗. This guide’s Homebrew commands are for macOS; ask Claude to adapt the tool setup for WSL rather than pasting them into PowerShell.
1START WITH CLAUDE CODE
Install Claude Code.
Open Terminal and review the official quickstart ↗. Its recommended native installer for macOS, Linux, and WSL is:
COMMAND · Install Claude Code
curl -fsSL https://claude.ai/install.sh | bash
Open a new terminal window after installation and check it works:
COMMAND · Verify Claude Code
claude --version
If the command is not found, follow the official PATH troubleshooting guide. First-time sign-in happens when you launch Claude.
2PUT THE TOOLS IN PLACE
Your local toolkit.
Install Homebrew ↗ first if it is missing. These commands create an isolated Python environment inside ReelLab:
Whisper runs locally after its model downloads. Faster-whisper is a Python library; Claude will build the wrapper that writes the transcript files. Public-post availability and metadata depend on the platform.
3LET CLAUDE BUILD IT
One prompt. A repeatable process.
Start Claude inside your workspace:
COMMAND · Open ReelLab in Claude
cd ~/ReelLab
source .venv/bin/activate
claude
Paste this setup prompt into Claude Code. Review its plan and approve the installations you want.
PROMPT · Let Claude build the workflow
Set up a reel-analysis workflow on this Mac.
1. Check for Homebrew, Python 3, ffmpeg, yt-dlp, instaloader and faster-whisper. Tell me what's missing and ask before installing anything.
2. Create a folder ~/ReelLab with reels/ and lessons/ inside.
3. Write a command `analyze <url>` that:
- downloads the video, caption and metadata into reels/<date>-<short-name>/ (yt-dlp first; instaloader for Instagram if yt-dlp fails)
- extracts a keyframe every second plus every scene change with ffmpeg, and makes one contact sheet image
- transcribes the audio with timestamps to transcript.txt and transcript.srt
- writes summary.json with the duration, number of cuts and words per minute
4. Test it on one public reel I'll give you, then show me what's in the folder.
Never log in to any account without asking me first.
What “analyze” means here
It is the custom command Claude creates in this step, not a command installed by yt-dlp or Instaloader. Ask Claude to confirm how to run it and which files it produced. Scene-cut counts and words per minute are estimates; check the implementation and sample output.
The prompt is a starting specification, not a prebuilt app. Ask Claude to show a successful test with a permitted public reel, mark unavailable metadata, and avoid inventing timestamps or visual details.
4STUDY WHAT’S ACTUALLY THERE
Find the useful patterns.
Replace reels/<folder> with the folder created by your first analysis. Give Claude the frames, transcript, and summary together.
PROMPT · Break down a reel
Break down this reel. What's the opening hook? How does the pacing change? How do the visuals support the message? And what's the call to action?
Use the keyframes, the transcript and summary.json in reels/<folder>. Quote timestamps for every point.
5TURN LESSONS INTO ORIGINAL WORK
Make it yours.
Add your audience, message, and offer after the prompt. Use the underlying principles with your own examples.
PROMPT · Make it yours
Use those principles to help me create an original reel for my audience, with my own examples and voice.
Don't copy the original's wording or visuals. Give me a hook, a beat-by-beat script with on-screen text, and a call to action.
6BUILD A LIBRARY OF LESSONS
Your saved links, connected.
Create links.txt in ReelLab with one permitted public URL per line. Work through them sequentially and keep the source folders.
PROMPT · Turn saved links into lessons
Run analyze on each link in links.txt, one at a time. Then compare all of the breakdowns and write lessons/lessons.md with:
the hook patterns that repeat, the pacing patterns, the visual techniques, and the CTAs, each with the reels that use them.
End with 5 ideas for my next video based on these lessons.
A local links file is enough; this does not require importing your private Instagram saved collection. Slow down or stop if the platform returns rate limits.
USE IT RESPONSIBLY
Study the principles. Create your own.
Download only material you are allowed to use, for your own study, and respect the platform’s terms.
Never repost someone else’s video. Use your own wording, examples, footage, and visuals.
Some Instagram posts require login. Use your own account only when appropriate, ask before logging in, and avoid repeated requests.
Private accounts and access-control workarounds are outside this guide.
Independent guide by Rami Raidan. No affiliation with Anthropic, Instagram/Meta, or the open-source tool authors.
WHEN SOMETHING GETS STUCK
A quick check.
yt-dlp returns an extraction error
Update the package in your ReelLab environment, then check the official project issues if the platform changed.
Instaloader returns 401 or 429
Stop repeated retries. Check your own session or login only when needed, and allow a cooldown; do not try to bypass limits.
“ffmpeg: command not found”
Confirm the Homebrew installation and open a new Terminal window. Check that Homebrew’s bin directory is on your PATH.
Whisper takes too long
Ask Claude to use a smaller model on CPU and test a short clip first. Let the first model download finish.