TRex - Text Recognition for Arch Linux

Lightning-fast OCR tool optimized for Wayland/Hyprland on Arch Linux. Select screen area → Extract text → Copy to clipboard.

Features

• 🎯 Simple - One command to capture and extract text
• ⚡ Lightning Fast - Tesseract by default, results in under 0.5s
• 🖥️ Wayland Native - Built for modern Linux desktops
• 📋 Clipboard Integration - Automatically copies extracted text
• 🎨 Flexible - Fast mode by default, --accurate flag for complex text

Installation

# Install system dependencies (Arch Linux)
sudo pacman -S grim slurp wl-clipboard tesseract tesseract-data-eng

# Install TRex (lightweight by default)
pip install trex-linux

By default, TRex uses Tesseract for blazing-fast OCR. For better accuracy on complex images, install the accurate mode dependencies.

Usage

# Default: Select area → OCR → Copy to clipboard
trex

# OCR image from clipboard
trex -c

# OCR image file
trex -f screenshot.png

# Output to stdout instead of clipboard
trex -o stdout

# Use different language (e.g., German)
trex -l deu

# Use accurate mode for complex/handwritten text
trex --accurate

Performance

• Default mode: ~0.3-0.5s (Tesseract)
• Accurate mode: ~2-3s (EasyOCR, downloads AI models on first use)

Languages

Tesseract (default): Uses standard 3-letter codes

• eng - English (default)
• deu - German
• fra - French
• spa - Spanish
• jpn - Japanese
• Install more: sudo pacman -S tesseract-data-[lang]

EasyOCR (--accurate mode):

• en, de, fr, es, ja, ko, ch_sim, etc.
• Full list: https://github.com/JaidedAI/EasyOCR#supported-languages

Configuration

Optional config file at ~/.config/trex/config.json:

{
  "language": "en",
  "gpu": false
}

Set "gpu": true if you have NVIDIA GPU with CUDA support.

Requirements

• Arch Linux
• Wayland compositor (tested on Hyprland)
• Python 3.8+
• CUDA (optional, for GPU acceleration)

Accurate Mode Setup

For complex images, handwriting, or multiple languages:

# Install accurate mode dependencies  
pip install trex-linux[accurate] --extra-index-url https://download.pytorch.org/whl/cpu

# Or with GPU support (NVIDIA CUDA)
pip install trex-linux[accurate,gpu]

Then use trex --accurate for better accuracy at the cost of speed.

License

MIT