instagram-reel-extractor

Instagram Reel Extractor

A Claude Code skill that extracts transcripts, metadata, music, comments, and key frames from Instagram Reels — so you can analyze, repurpose, and build on content without manually rewatching videos.

The Problem

You find an Instagram Reel with a great script, a trending audio, or a content format you want to study. But to actually use any of it, you need to:

• Rewatch the reel multiple times to manually transcribe what was said
• Shazam the audio separately to figure out the background music
• Screenshot individual frames to reference the visual style or text overlays
• Copy-paste the caption and manually extract hashtags
• Scroll through comments to understand what resonated with the audience

This is tedious, slow, and doesn’t scale. If you’re a content creator, marketer, or strategist studying what works on Instagram, you’re spending more time extracting information than actually using it.

The Solution

Paste a Reel URL and get everything extracted into a single structured markdown file in seconds:

/instagram-reel-extractor https://www.instagram.com/reel/ABC123/

The skill downloads the video, transcribes the audio with AI, identifies the background music, pulls all metadata, extracts key frames, and saves it all as a clean markdown document you can reference, search, and repurpose.

What Gets Extracted

Feature	How It Works
Full transcript	OpenAI Whisper (runs locally, free, no API key)
Timestamped segments	Every sentence mapped to its timecode
Music identification	Shazam audio fingerprinting with song, artist, genre, and link
Creator info	Display name, handle, profile ID
Engagement metrics	Likes, comments count (as of extraction date)
Caption & hashtags	Full original caption with hashtag list
Top comments	Comment text, author, and like counts
Key frames	Screenshots every N seconds (configurable)
Original URL	Always linked at the top for reference

Use Cases

• Content repurposing — Extract scripts from viral reels to adapt for your own content, blog posts, tweet threads, or YouTube shorts
• Competitor research — Study what top creators in your niche are saying, what music they use, and what hooks get engagement
• Trend tracking — Build a library of reel extractions in Obsidian, Notion, or any markdown-based system to spot patterns over time
• Music discovery — Find trending audio for your own reels without manually Shazaming
• Comment analysis — See what questions and reactions the audience has to inform your content strategy
• Visual reference — Key frames show the editing style, text overlays, and B-roll approach without rewatching

Install

npx skills add JayceBatallones/instagram-reel-extractor -g

Dependencies

Run the setup script:

bash setup.sh

Or install manually:

brew install yt-dlp ffmpeg
pip3 install openai-whisper ShazamAPI

Note: OpenAI Whisper is completely free and open-source. It runs locally on your machine — no API key or OpenAI account needed. It’s not the same as OpenAI’s paid Whisper API.

Usage

Basic extraction

Just paste a Reel URL in Claude Code:

Extract this reel: https://www.instagram.com/reel/ABC123/

Or invoke the skill directly:

/instagram-reel-extractor https://www.instagram.com/reel/ABC123/

Save to a folder

Perfect for building a content library in Obsidian, Notion, or any notes app:

Extract this reel and save to ~/obsidian/content-ideas/

Extractions are saved as <creator>-reel-<date>.md (e.g., garyvee-reel-2026-04-08.md). Duplicates are auto-numbered.

Fast mode (skip frames)

If you only need the transcript and metadata:

Extract this reel with no frames: https://www.instagram.com/reel/ABC123/

Private reels

If a reel requires login, pass your browser cookies:

python3 scripts/extract_reel.py "URL" --cookies-from chrome

Options

Flag	Description
--save-dir PATH	Save extraction markdown to this directory
--whisper-model MODEL	Whisper model: tiny, base (default), small, medium, large
--frame-interval N	Seconds between frame captures (default: 2.0)
--no-frames	Skip frame extraction (faster)
--cookies-from BROWSER	Use browser cookies for auth (e.g., chrome, firefox)
--json	Output raw JSON instead of markdown

Whisper Model Sizes

Model	Speed	Accuracy	Memory
tiny	Fastest	Lower	~1 GB
base	Fast	Good	~1 GB
small	Moderate	Better	~2 GB
medium	Slow	Great	~5 GB
large	Slowest	Best	~10 GB

Example Output

# Instagram Reel Extraction

**Source:** [https://instagram.com/reel/ABC123/](...)
**Extracted:** 2026-04-08 08:15 UTC

## Metadata
- **Creator:** Gary Vaynerchuk (@garyvee)
- **Duration:** 0:58
- **Likes:** 45,231
- **Comments:** 1,204

## Music
- **Song:** Runaway (feat. Pusha T)
- **Artist:** Kanye West
- **Shazam:** https://www.shazam.com/track/52689669/runaway-feat-pusha-t

## Caption
Stop overthinking. Start posting. The algorithm rewards consistency, not perfection...

## Top Comments
*48 comments extracted*
- **@entrepreneur.daily** (12 likes): This changed my entire content strategy
- **@sarah_creates**: How often do you post reels vs stories?

## Transcript
*Language detected: en*

The number one mistake I see creators make is they wait for
the perfect idea. There is no perfect idea. The market decides
what's perfect, not you...

## Timestamped Transcript
[00:00] The number one mistake I see creators make...
[00:04] is they wait for the perfect idea...
[00:08] There is no perfect idea. The market decides...

How It Compares

This skill is built for deep content analysis of individual reels. If you need bulk data scraping (profiles, followers, hashtag search), tools like Apify’s Instagram Scraper are better suited. Here’s the tradeoff:

	instagram-reel-extractor	Bulk scraping tools
Cost	Free (all open-source)	Paid (API subscriptions)
Transcript	Yes	No
Music ID	Yes	No
Key frames	Yes	No
Runs locally	Yes (private)	No (cloud)
Bulk scraping	No	Yes
Profile/follower data	No	Yes

Troubleshooting

• yt-dlp fails: Make sure yt-dlp is up to date (brew upgrade yt-dlp). Some reels may require --cookies-from chrome.
• No speech detected: The reel may be music-only or use on-screen text. The music identification and frame extraction still work.
• Slow transcription: Use --whisper-model tiny for faster results, or small for better accuracy.
• Music not identified: Shazam works best with distinct background music. Speech-heavy reels without music will return no match.

License

MIT