Loyal Bear – The SynthID Scrambler
Purpose of this Project
Corporations continue to exert control over AIs under the guise of "safety". Their recent intrusion into the AI imaging field is the mandatory incorporation of "SynthID". This is applied without consent and cannot be opted out of, even by paying users. It works by hiding a pattern of pixels within the image, not noticeable by the human eye but instead functioning as an invisible watermark. Additional information could be hidden within the watermark, much like a QR code. This pattern can be used to track and deanonymize users.
Upon discovering that other SynthID removal tools do not actually fix this, I created my own. It works on all OpenAI and Gemini images as of July 2026, scrambling all trackers and watermarks while doing minimal damage to the image.
I invite all individuals to use it for personal and commercial use. Corporations and individuals acting on behalf of corporations are strictly forbidden from using or examining this tool. I've incorporated a system that alerts me to compromise attempts.
To preserve the longevity of this scrambler, I will be keeping most of its methods secret. What I can tell you is that it will run on any computer with Python 3.10+ and 8GB of RAM. The Windows version has been thoroughly tested, while the Linux version has not. Please open a bug report if you encounter issues.
If this project is successful, I'll make something similar for text SynthIDs.
Quick Start
Windows — double-click run.bat
Linux/macOS — chmod +x run.sh && ./run.sh
The first launch will install Python dependencies, download the model (~6.9 GB), and open the application. Subsequent launches start instantly.
Developer Setup
python -m venv .venv
.venv\Scripts\activate # Windows
# source .venv/bin/activate # Linux/macOS
pip install -r requirements.txt
python main.py
Building for distribution
python build_release.py # compile backend to .pyd
python build_release.py --restore # restore source for development
Requirements
- Python 3.10+
License
See LICENSE