TVS - Text-Video-Summarizer

A standalone Python tool for downloading, transcribing, and summarizing videos automatically.

Features

✅ Single Video Processing - Process one video at a time ✅ Batch Processing - Process multiple videos from a URL list file ✅ Smart Download Detection - Detects already downloaded videos and skips re-downloading ✅ Vibe-Only Transcription - Uses only vibe (Whisper Large V3 Turbo) for accurate transcription ✅ Automatic Summarization - Generates structured markdown summaries ✅ Colored Terminal Output - Beautiful progress indicators and status messages ✅ Error Handling - Clear error messages with troubleshooting steps

Requirements

• Python 3.13+
• yt-dlp (video downloader)
• vibe (AI transcription tool)
• Whisper Large V3 Turbo model (1.5 GB, auto-downloaded by vibe GUI)

Installation

# Install yt-dlp
sudo pacman -S yt-dlp

# Install vibe
yay -S vibe-bin

# Run vibe GUI once to download the model
vibe

# Make script executable
chmod +x tvs.py

Usage

Single Video

python3.13 tvs.py -u <URL>
python3.13 tvs.py --url <URL>

Example:

python3.13 tvs.py -u https://vediosite.com/123

Batch Processing

python3.13 tvs.py -l <file.txt>
python3.13 tvs.py --list <file.txt>

Example:

python3.13 tvs.py --list my-videos.txt

URL List File Format

# Comments start with #
# Blank lines are ignored

https://vediosite.com/123
https://vediosite.com/123

# Another comment
https://youtu.be/VIDEO3

Help

python3.13 tvs.py --help

Output Structure

Files Created

~/Videos/
├── Video_Title_Here.webm                    # Downloaded video
└── Video_Title_Here-transcript.txt          # Transcription

~/Work/Kai/video/
├── Video_Title_Here-transcript.txt          # Transcription (backup)
└── Video_Title_Here-summarize.md            # Summary ⭐

Summary Contents

Each generated summary includes:

• Executive Summary
• Key Takeaways
• Full Transcript
• Analysis Notes (word count, sentence count, etc.)
• Processing Details (workflow, file locations, tools used)

Processing Time

For a typical 1-minute video:

• Download: ~5-15 seconds
• Transcription: ~10-15 seconds
• Summary generation: ~2-5 seconds
• Total: ~20-30 seconds

Features in Detail

1. Smart Download Detection

If a video was already downloaded:

⚠️  Video already exists: Video_Title_Here.webm
ℹ️  Skipping download (file already present)

The script will skip the download and proceed directly to transcription.

2. Batch Processing Statistics

After processing multiple videos:

Batch Processing Complete!

Statistics:
  ✅ Successful: 5
  ❌ Failed:     1
  📊 Total:      6
  ⏱️  Time:       125.3s (2.1 min)

3. Error Handling

If transcription fails, you get actionable troubleshooting steps:

❌ Transcription failed
⚠️  Troubleshooting:
  1. Verify model exists:
     ls -lh ~/.local/share/github.com.thewh1teagle.vibe/ggml-large-v3-turbo.bin
  2. Check video has audio:
     ffprobe -v error -select_streams a:0 ...
  3. Try running vibe GUI first

Command Reference

Command	Description
python3.13 tvs.py -u <URL>	Process single video
python3.13 tvs.py --url <URL>	Process single video (long form)
python3.13 tvs.py -l <file>	Process URLs from file
python3.13 tvs.py --list <file>	Process URLs from file (long form)
python3.13 tvs.py --help	Show help message

Configuration

Directories

Configured at the top of the script:

VIDEOS_DIR = Path.home() / "Videos"
WORK_DIR = Path.home() / "Work" / "Kai" / "video"

Vibe Model

VIBE_MODEL = Path.home() / ".local/share/github.com.thewh1teagle.vibe/ggml-large-v3-turbo.bin"

Timeouts

# In run_command() function
timeout=600  # Download timeout: 10 minutes
timeout=900  # Transcription timeout: 15 minutes

Workflow

1. Environment Validation
   ├── Check yt-dlp installed
   ├── Check vibe installed
   ├── Check vibe model exists
   └── Create directories if needed

2. Download Video (yt-dlp)
   ├── Check if already exists
   ├── Download with --restrict-filenames
   └── Verify video file created

3. Transcribe Audio (vibe)
   ├── Use Whisper Large V3 Turbo model
   ├── Automatic language detection
   └── Save as plain text

4. Copy Transcript
   └── Create backup in work directory

5. Generate Summary
   ├── Read transcript
   ├── Analyze content
   ├── Create structured markdown
   └── Save to work directory

Troubleshooting

Model Not Found

# Run vibe GUI to download model
vibe

The model will be downloaded to:

~/.local/share/github.com.thewh1teagle.vibe/ggml-large-v3-turbo.bin

Video Has No Audio

Some videos don't have audio tracks. Check with:

ffprobe -v error -select_streams a:0 -show_entries stream=codec_name VIDEO_FILE

Download Fails

• Check internet connection
• Verify URL is valid
• Try updating yt-dlp: yay -S yt-dlp

Transcription Timeout

For very long videos (>1 hour), you may need to increase the timeout in the script:

success, stdout, stderr = run_command(cmd, cwd=VIDEOS_DIR, timeout=1800)  # 30 min