Reverse Engineering Google's SynthID Watermark: Detection and Removal Through Spectral Analysis
By
_tk_
Pure flour-power. Hearty enough to carry you through lunch.
Summary
This article describes a GitHub project that reverse-engineers Google's SynthID watermarking system used in Gemini-generated images. The project uses signal processing and spectral analysis to discover the watermark's resolution-dependent carrier frequency structure, build a detector with 90% accuracy, and develop a multi-resolution spectral bypass that achieves 75% effectiveness in removing the watermark without access to Google's proprietary encoder/decoder.
Key quotes
· 5 pulledThis project reverse-engineers Google's SynthID watermarking system - the invisible watermark embedded into every image generated by Google Gemini
Using only signal processing and spectral analysis (no access to the proprietary encoder/decoder)
Discovered the watermark's resolution-dependent carrier frequency structure
Built a detector that identifies SynthID watermarks with 90% accuracy
Developed a multi-resolution spectral bypass (V3) that achieves 75% c
You might also wanna read
Developer Claims to Reverse-Engineer Google's AI Watermarking System
A software developer claims to have reverse-engineered Google DeepMind's SynthID AI watermarking system, demonstrating how to strip watermar
The Verge·1mo ago
Google Gemini Adds AI-Generated Image Detection Feature
Google is introducing a new feature in its Gemini app that allows users to detect AI-generated images, but only those created or edited by G
The Verge·6mo ago
Google Gemini App Adds Video AI Verification Feature
Google has expanded its AI verification feature in the Gemini app to include video content. The feature allows users to upload videos and as
The Verge·5mo ago
Google expands deepfake detection to Chrome and Search with SynthID and C2PA support
Google is expanding its AI detection capabilities to Chrome and Search, making it easier for users to identify deepfakes. The updates includ
The Verge·12d ago
Google expands SynthID and C2PA content credentials to combat AI-generated deepfakes
Google's SynthID and the C2PA Content Credentials system are undergoing their largest expansion yet, aiming to combat AI-generated deepfakes
The Verge·11d ago