TVS - Text-Video-Summarizer

A standalone Python CLI tool for downloading, transcribing, and summarizing videos from social media platforms.

Features

Core Features

Single Video Processing - Process one video at a time
Batch Processing - Process multiple videos from a URL list file
Smart Caching - Skips already downloaded videos and existing transcripts
Audio-Only Mode - Download only audio for faster processing (-a flag)
Live Terminal Output - Real-time progress updates (-t flag)
Parakeet Transcription - NVIDIA NeMo parakeet-tdt-0.6b-v3 for accurate ASR
AI Summarization - OpenCode agent generates intelligent summaries
Color Demo - View available terminal colors (--colors flag)

v4.0 Features (March 2026)

Parakeet Migration - Replaced vibe with NVIDIA NeMo parakeet-tdt-0.6b-v3
Cross-Platform Support - macOS, Ubuntu/Debian, Fedora, Arch Linux
Local Model - 2.3 GB model stored in repository directory
Conda Environment - Isolated nemo environment for transcription
Opencode-Style Colors - Clean, semantic color names

v3.x Features

Multi-Site Support - YouTube, Instagram, TikTok, X/Twitter
Cookie Management - Organized by site with expiration warnings
AI Smart Naming - Descriptive filenames for social media videos
Automatic Hashtags - 3-5 relevant hashtags in summaries
Platform Detection - Auto-detect source and organize accordingly

Requirements

Requirement	Purpose	Install
Python 3.13+	Runtime	System package
yt-dlp	Video downloading	`brew install yt-dlp` / `pip install yt-dlp`
ffmpeg	Audio conversion	`brew install ffmpeg` / `sudo apt install ffmpeg`
mediainfo	Duration detection (optional)	`brew install mediainfo`
conda (miniforge)	Environment management	`brew install --cask miniforge`
nemo_toolkit	ASR transcription	`pip install 'nemo_toolkit[asr]'`
opencode	Summary generation	See opencode.ai

Installation

macOS (Apple Silicon/Intel)

# Install Homebrew if not installed
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Install dependencies
brew install yt-dlp ffmpeg mediainfo

# Install Miniforge (conda)
brew install --cask miniforge

# Initialize conda
conda init zsh  # or bash
source ~/.zshrc

# Create nemo environment
conda create -n nemo python=3.11 -y
conda activate nemo
pip install 'nemo_toolkit[asr]'

# Clone repository
git clone https://github.com/commandlinetips/tvs.git
cd tvs

# Download parakeet model (2.3 GB)
git clone https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3

# Verify model
ls -la parakeet-tdt-0.6b-v3/parakeet-tdt-0.6b-v3.nemo

Ubuntu/Debian

# Install system dependencies
sudo apt update
sudo apt install -y python3.11 python3-pip ffmpeg mediainfo

# Install yt-dlp
pip install yt-dlp --break-system-packages
# Or: sudo wget https://github.com/yt-dlp/yt-dlp/releases/latest/download/yt-dlp -O /usr/local/bin/yt-dlp && sudo chmod +x /usr/local/bin/yt-dlp

# Install Miniforge
wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh
bash Miniforge3-Linux-x86_64.sh
source ~/.bashrc

# Create nemo environment
conda create -n nemo python=3.11 -y
conda activate nemo
pip install 'nemo_toolkit[asr]'

# Clone repository
git clone https://github.com/commandlinetips/tvs.git
cd tvs

# Download parakeet model (2.3 GB)
git clone https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3

Fedora

# Install system dependencies
sudo dnf install -y python3.11 python3-pip ffmpeg mediainfo

# Install yt-dlp
pip install yt-dlp

# Install Miniforge
wget https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh
bash Miniforge3-Linux-x86_64.sh
source ~/.bashrc

# Create nemo environment
conda create -n nemo python=3.11 -y
conda activate nemo
pip install 'nemo_toolkit[asr]'

# Clone and setup
git clone https://github.com/commandlinetips/tvs.git
cd tvs
git clone https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3

Arch Linux

# Install system dependencies
sudo pacman -S yt-dlp ffmpeg mediainfo python

# Install miniforge from AUR
yay -S miniforge

# Initialize and create environment
conda init zsh
source ~/.zshrc
conda create -n nemo python=3.11 -y
conda activate nemo
pip install 'nemo_toolkit[asr]'

# Clone and setup
git clone https://github.com/commandlinetips/tvs.git
cd tvs
git clone https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3

Usage

Quick Start

# Single video - audio-only with live output (recommended)
python3.13 tvs.py -u "VIDEO_URL" -a -t

# Force re-transcription
python3.13 tvs.py -u "VIDEO_URL" -a -t -f

# Download only (skip transcription)
python3.13 tvs.py -u "VIDEO_URL" -a -t -d

# Batch processing
python3.13 tvs.py --list urls.txt -a -t

# Show color demo
python3.13 tvs.py --colors

Command Flags

Flag	Long Form	Description
`-u`	`--url`	Single video URL
`-l`	`--list`	URL list file (one per line)
`-a`	`--audio-only`	Download audio only (faster, smaller)
`-t`	`--terminal`	Show live output (no buffering)
`-f`	`--force`	Force re-transcription
`-d`	`--download-only`	Download only, skip transcription
	`--colors`	Show color demo and exit
`-h`	`--help`	Show help message

URL List Format

# Comments start with #
# Blank lines are ignored

https://youtube.com/watch?v=ABC123
https://instagram.com/reel/XYZ789

# More URLs
https://tiktok.com/@user/video/123456

Directory Structure

tvs/
├── tvs.py                    # Main script
├── videos/                   # Downloaded media
│   ├── youtube/
│   ├── instagram/
│   ├── tiktok/
│   └── x/
├── transcripts/              # Transcripts and summaries
├── logs/                     # Debug and processed logs
├── cookies/                  # Platform cookies (optional)
│   ├── instagram/
│   ├── tiktok/
│   └── x/
└── parakeet-tdt-0.6b-v3/     # Transcription model (2.3 GB)
    └── parakeet-tdt-0.6b-v3.nemo

Processing Pipeline

URL → download_video() → transcribe_video() → copy_transcript() → generate_summary()
        (yt-dlp)            (parakeet)           (shutil)          (OpenCode)

Step Details

Environment Validation - Check yt-dlp, ffmpeg, conda, nemo env
Download - yt-dlp downloads video/audio to videos/<platform>/
Transcribe - parakeet-tdt-0.6b-v3 transcribes audio (requires conda env)
Copy - Transcript copied to transcripts/
Summarize - OpenCode agent generates AI summary

Output Files

videos/
├── youtube/Video_Title.m4a           # Audio file
├── youtube/Video_Title-transcript.txt # Transcript

transcripts/
├── Video_Title-transcript.txt        # Transcript copy
└── Video_Title-summarize.md          # AI summary with hashtags

Platform-Specific Naming

Platform	Video Filename	Summary Filename
YouTube	`Video_Title.m4a`	`Video_Title-summarize.md`
Instagram	`C_XYZ123.m4a`	`ai-topic-C_XYZ123-summarize.md`
TikTok	`7123456789.m4a`	`ai-topic-7123456789-summarize.md`
X/Twitter	`username_123.m4a`	`username_123-summarize.md`

Processing Times

Audio-Only Mode (Recommended)

Video Length	Download	Transcribe	Summary	Total
1 minute	~5s	~10s	~30s	~45s
10 minutes	~15s	~1m	~30s	~2m
60 minutes	~60s	~5m	~45s	~7m

Why Audio-Only?

10x smaller files (50 MB vs 500 MB)
10x faster downloads
Transcription only needs audio
Same quality transcripts

Cookie Setup (Social Media)

For Instagram, TikTok, and X/Twitter, cookies may be required.

Export Cookies

Install browser extension "Get cookies.txt LOCALLY"
Log into the platform in your browser
Export cookies for the site

Cookie Locations

cookies/
├── instagram/www.instagram.com_cookies.txt
├── tiktok/www.tiktok.com_cookies.txt
├── x/x.com_cookies.txt
└── youtube/www.youtube.com_cookies.txt

Cookie Warnings

TVS warns when cookies are 30+ days old:

⚠️  Cookie file is 32 days old (threshold: 30 days)
⚠️  COOKIES MAY HAVE EXPIRED - CONSIDER UPDATING

Troubleshooting

Model Not Found

# Verify model exists
ls -la parakeet-tdt-0.6b-v3/parakeet-tdt-0.6b-v3.nemo

# Download if missing
git clone https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3

Conda Environment Issues

# Verify nemo environment
conda env list

# Recreate if needed
conda create -n nemo python=3.11 -y
conda activate nemo
pip install 'nemo_toolkit[asr]'

# Test import
python -c "import nemo.collections.asr; print('OK')"

Transcription Fails

# Check video has audio
ffprobe -v error -select_streams a:0 VIDEO_FILE

# Check ffmpeg
ffmpeg -version

# Run manually to debug
conda activate nemo
python -c "
import nemo.collections.asr as nemo_asr
model = nemo_asr.models.ASRModel.restore_from('parakeet-tdt-0.6b-v3/parakeet-tdt-0.6b-v3.nemo')
print(model.transcribe(['test.wav']))
"

Download Fails

Check internet connection
Verify URL is valid
Update yt-dlp: yt-dlp -U or pip install -U yt-dlp
For social media: ensure cookies are fresh

Threads Not Supported

yt-dlp doesn't support Threads yet. Workaround:

Download Threads video manually
Place in videos/threads/
Transcription and summary will work

Supported Platforms

Platform	Download	Transcribe	Summary	AI Naming	Status
YouTube	Yes	Yes	Yes	No	Fully working
Instagram	Yes	Yes	Yes	Yes	Fully working
TikTok	Yes	Yes	Yes	Yes	Fully working
X/Twitter	Yes	Yes	Yes	No	Fully working
Threads	No	Yes*	Yes*	Yes*	Manual download

*Requires manual video download first

Model Information

parakeet-tdt-0.6b-v3

Source: NVIDIA NeMo
Size: 2.3 GB
Download: https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
Requirements: nemo_toolkit[asr], Python 3.11+

# Clone model
git clone https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3

# Or download specific file
wget https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/resolve/main/parakeet-tdt-0.6b-v3.nemo

Version History

v4.0 (March 2026)

Migrated from vibe to parakeet-tdt-0.6b-v3 (NVIDIA NeMo)
Added cross-platform installation instructions
Added conda environment isolation (nemo)
Updated color class to opencode-style semantic names
Added --colors flag for color demo
Model stored in repository directory (not user home)
Removed dependency on vibe-bin

v3.1 (November 2025)

Multi-site support with platform detection
Organized cookie management by site
Cookie expiration warnings (30-day threshold)
AI-powered smart naming for social media
Automatic hashtag generation

v3.0 (October 2025)

Smart transcript caching
AI summaries via OpenCode
Audio-only mode
Live terminal output
Force re-transcription flag

v2.0 (Earlier)

Batch processing support
Smart download detection
Colored terminal output

License

MIT License

Links

GitHub: https://github.com/commandlinetips/tvs
Model: https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
NeMo: https://github.com/NVIDIA/NeMo
OpenCode: https://opencode.ai

Created: 2025-10-31 Updated: 2026-03-27 Author: skullthoughts Version: TVS v4.0

Name		Name	Last commit message	Last commit date
Latest commit History 14 Commits
docs/superpowers		docs/superpowers
.gitignore		.gitignore
AGENTS.md		AGENTS.md
GEMINI.md		GEMINI.md
README.md		README.md
USAGE.md		USAGE.md
test-urls.txt		test-urls.txt
tvs.py		tvs.py

Folders and files

Latest commit

History

Repository files navigation

TVS - Text-Video-Summarizer

Features

Core Features

v4.0 Features (March 2026)

v3.x Features

Requirements

Installation

macOS (Apple Silicon/Intel)

Ubuntu/Debian

Fedora

Arch Linux

Usage

Quick Start

Command Flags

URL List Format

Directory Structure

Processing Pipeline

Step Details

Output Files

Platform-Specific Naming

Processing Times

Audio-Only Mode (Recommended)

Why Audio-Only?

Cookie Setup (Social Media)

Export Cookies

Cookie Locations

Cookie Warnings

Troubleshooting

Model Not Found

Conda Environment Issues

Transcription Fails

Download Fails

Threads Not Supported

Supported Platforms

Model Information

parakeet-tdt-0.6b-v3

Version History

v4.0 (March 2026)

v3.1 (November 2025)

v3.0 (October 2025)

v2.0 (Earlier)

License

Links

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages