PixelPhantomX: Ghost images to confuse AI models

PixelPhantomX is a powerful image obfuscation tool designed to generate ghost images that confuse AI models. In an era where generative AI scrapes and repurposes artwork without the consent of creators, this tool acts as a defensive mechanism. By applying various adversarial techniques, such as noise injection, color-space manipulation, edge distortion, and watermarking, PixelPhantomX helps protect digital images from unauthorized AI training and exploitation.

Features

• Adversarial Noise Injection (Inspired by Glaze & Nightshade)
• Customizable Noise Levels (Min/Max)
• Watermarking (Visible & Invisible Steganographic Protection)
• Edge Distortion for AI Confusion
• Metadata Poisoning to mislead AI training
• Command-line Interface for easy use

Installation

Direct Install from PyPI

pip install PixelPhantomX

Alternatively, install via setup.py:

python setup.py install

Otherwise, install the required dependencies:

pip install -r requirements.txt

Usage

Run the tool via the command line:

python PixelPhantomX.py -iP input_image_path -n noise -oP output_image_path -w watermark_text

Parameters:

Example:

python PixelPhantomX.py -iP Input.png -oP Output -n Min -w AdityaBhatt3010

Help Menu

To view available commands, run:

python PixelPhantomX.py -h

Alternative Usage - Interactive Mode

For a guided, interactive experience instead of command-line arguments, use:

python PixelPhantomX_Minimal.py

Screenshots

• Help Prompt Screenshot

  python PixelPhantomX.py -h
  

  
  

• Run Code Example:

  python PixelPhantomX.py -iP Input.png -oP Output -n Min -w AdityaBhatt3010
  

  
  

• Input Image Screenshot
  
  

• Output Image Screenshot
  
  

License

This project is licensed under the MIT License.

Author

Aditya Bhatt\

Contributing

Pull requests are welcome! For major changes, please open an issue first to discuss what you'd like to change.

Disclaimer:

This tool is intended for ethical use only. The author is not responsible for any misuse of this software.