YouTube Ripper

Local web utility that extracts audio and video from YouTube, transcribes with Whisper, and generates clean text transcripts.

Features

• Five output modes: Audio Only, Text Only, Audio + Text, Video, Video + Text
• Video quality selection: 480p, 720p, 1080p, or Best Available
• Audio format: M4A (AAC) — plays on Mac, Android, iOS
• Video format: MP4 with merged audio
• Transcription: Local Whisper (faster-whisper) — no cloud APIs, no data leaves your machine
• Text output: Clean .txt transcript with title, source URL, and paragraph breaks
• Real-time progress via Server-Sent Events (download %, speed, ETA for video)
• Supported URLs: youtube.com/watch, youtu.be, youtube.com/shorts, youtube.com/live

Output Modes

Mode	What you get	Quality picker?
Audio Only	M4A	No
Text Only	TXT transcript	No
Audio + Text	M4A + TXT	No
Video	MP4	Yes
Video + Text	MP4 + TXT	Yes

Video + Text downloads the video, extracts the audio track, runs Whisper on it, generates the text transcript, and cleans up the intermediate audio file.

Requirements

• Python 3.12+
• uv package manager
• ffmpeg (brew install ffmpeg)

Quick Start

cd tools/youtube-ripper
uv run python app.py

Open http://localhost:4039, paste a YouTube URL, pick your mode, and hit Rip It.

For development with auto-reload:

FLASK_DEBUG=1 uv run python app.py

Stack

Component	Tool
YouTube download	yt-dlp
Transcription	faster-whisper (base model, int8)
Text generation	Python pathlib
Web server	Flask
Frontend	Vanilla HTML/CSS/JS

Output

Files are saved to output/ by default (gitignored). The UI shows clickable saved-file links after each run, so you can download the generated files directly from the browser.

Generated filenames include the YouTube id to avoid collisions between same-title videos:

• Title [video_id].m4a
• Title [video_id].mp4
• Title [video_id].txt

In "Text Only" mode, the downloaded audio file is automatically deleted after transcription. In "Video + Text" mode, the app downloads the video once, extracts temporary transcription audio locally with ffmpeg, and removes that temporary audio after transcription.

Notes

• First run downloads the Whisper base model (~150MB)
• No database, no accounts, no cloud services
• Port: 4039
• The server binds to 0.0.0.0 for Tailscale reachability, but requests are accepted only from loopback and Tailscale client ranges by default.
• YouTube downloads use your local Chrome Profile 1 cookies by default to avoid bot/sign-in checkpoints.
• To override auth, set YOUTUBE_RIPPER_COOKIES_FROM_BROWSER, for example firefox:default, or set YOUTUBE_RIPPER_COOKIES to a Netscape-format cookies file.

Configuration

Optional environment variables:

Variable	Default	Purpose
YOUTUBE_RIPPER_OUTPUT_DIR	output/	Directory for generated files
YOUTUBE_RIPPER_HOST	0.0.0.0	Flask bind host
YOUTUBE_RIPPER_PORT	4039	Flask port
YOUTUBE_RIPPER_ALLOWED_CLIENTS	127.0.0.0/8,::1/128,100.64.0.0/10,fd7a:115c:a1e0::/48	Comma-separated allowed client IPs/CIDR ranges
YOUTUBE_RIPPER_WHISPER_MODEL	base	faster-whisper model name
YOUTUBE_RIPPER_WHISPER_DEVICE	cpu	faster-whisper device
YOUTUBE_RIPPER_WHISPER_COMPUTE_TYPE	int8	faster-whisper compute type
YOUTUBE_RIPPER_COOKIES_FROM_BROWSER	chrome:Profile 1	Browser cookies source for yt-dlp
YOUTUBE_RIPPER_COOKIES	unset	Netscape-format cookies file path