Developer Open-Sources Tool to Analyze and Challenge Google DeepMind's SynthID AI Watermarks

# Software Developer Open-Sources Method to Strip Google DeepMind SynthID Watermarks from AI Images A software developer using the username Aloshdenny has open-sourced work on GitHub that reverse-engineers Google DeepMind’s SynthID system. Aloshdenny claims to have shown how AI watermarks can be stripped from generated images or manually inserted into other works.

The Verge reported on the development in an article written by Jess Weatherbed and published on April 14, 2026, at 1:53 PM UTC.

Aloshdenny documented the process on Medium. The method required 200 Gemini-generated images, signal processing, and free time.

SynthID System Overview SynthID is a near-invisible watermarking system that tags content generated by Google’s AI tools.

It embeds itself in the pixels of images at the point of creation. SynthID was designed to be difficult to remove without degrading the image quality. SynthID is used across Google’s AI products, including models like Nano Banana and Veo 3.

It is applied to YouTube’s AI-generated creator clones.

Details of Aloshdenny's Process Aloshdenny's process to crack SynthID involves generating 200 entirely black or pure white images using Gemini.

It includes enhancing contrast and saturation, then denoising the saturation to expose watermark patterns. The process involves averaging the patterns to find the magnitude and phase of the watermark signal at every frequency bin per channel. Aloshdenny's process also involves hunting for signs of these frequencies in images and partially removing them at the angle of insertion.

' Aloshdenny was unable to remove SynthID entirely in tests and relied on confusing SynthID decoders.

Google's Response Google spokesperson Myriam Khan told The Verge: 'It is incorrect to say this tool can systematically remove SynthID watermarks.

' The Verge reported these statements in its coverage. Aloshdenny claims the work demonstrates vulnerabilities in watermark detection without full removal. The open-sourced code is available on GitHub for further examination.