Google's Gemini can now erase watermaks, and it's worryingly good at it

A week ago, Google released the new Gemini 2.0 Flash model’s image generation feature. With this new addition, the model natively generates and edits image content, which is great for those looking for a more powerful image generator.

But in the process, Google’s newest AI model seems to have received another unwanted side-feature: it can now remove watermarks from pictures, eliciting widespread alarm among content producers and copyright holders.

People have demonstrated that if they upload a photo with a watermark into the Gemini 2.0 Flash model and instruct it to remove the watermark, the AI deletes the mark, most often without leaving any visible evidence of alteration. This feature has sparked debates on possible misuse, particularly of copyright violation and misuse of licensed materials.

While this capacity illustrates AI’s sophisticated potential to handle image manipulation, it also reflects the ethical and legal concerns that accompany such technology. Intellectual property owners and sites relying on watermarks to protect their assets may have to reconsider their strategy in light of these developments.

With the development of AI technology, developers, users, and policymakers must engage in discussions about responsible use and implications for safeguarding digital content.