A software developer operating under the username Aloshdenny has published documentation claiming to reverse-engineer Google DeepMind's SynthID watermarking system, a technique used to mark AI-generated images. Google has disputed the effectiveness of the claim, describing it as incorrect.
SynthID is a near-invisible watermarking system that embeds itself in the pixels of images at the point of creation. It is used across Google's AI products, including tools like Gemini, Nano Banana, and Veo 3, and is also being applied to YouTube's AI-generated creator clones. The system was designed to be difficult to remove without degrading image quality.
Aloshdenny's approach, documented on GitHub and Medium, relies on signal processing rather than neural networks or proprietary access. The process involves generating 200 entirely black or pure white images using Gemini, then enhancing contrast and saturation before denoising to expose watermark patterns. These patterns are averaged to identify the magnitude and phase of the watermark signal at every frequency bin per channel. The final step involves hunting for signs of these frequencies in images and partially removing them at the same angle at which they were inserted during generation.
In his own testing, Aloshdenny acknowledged that he could not fully remove SynthID watermarks. Instead, his method can only confuse the decoders enough to cause them to fail detection. He described the underlying system as "genuinely good engineering" and noted that the inability to completely strip watermarks speaks to how effectively it was designed. "It's not perfect. But it's not trying to be unbreakable," he stated. "It's trying to raise the cost of misuse high enough that most people don't bother."
Google has formally rejected Aloshdenny's claims. Myriam Khan, a Google spokesperson, told The Verge: "It is incorrect to say this tool can systematically remove SynthID watermarks. SynthID is a robust, effective watermarking tool for AI-generated content."
The Verge's analysis indicates that, at this point, SynthID has not been reverse-engineered to a degree where widespread, simple tools could enable users to remove or add Google's watermarks to deceive AI detection systems. The practical risk remains limited, and the watermarking system continues to function as intended for the majority of use cases.