A controversial application of Google, the new Gemini AI Model has been identified by social media users and found to be able to remove watermarks from images, including those published by Getty Images and other prominent stock media companies.

Recently, Google expanded the access to the image generation feature of its Gemini 2.0 Flash model, which enables the model to natively generate and edit image content.

By all accounts, it is a potent capability. However, it also seems to have a limited number of guardrails.

Gemini 2.0 Flash will generate images that feature celebrities and copyrighted characters without hesitation, and, as previously mentioned, it will automatically eliminate watermarks from existing photographs.

Gemini 2.0 Flash will not merely eliminate watermarks; it will also endeavor to fill in any spaces that may have been left by the removal of a watermark, as several X and Reddit users have observed.

Other AI-powered tools also perform this function; however, Gemini 2.0 Flash appears to be particularly adept at it, and it is available for free.

The image generation feature of Gemini 2.0 Flash is currently designated as “experimental” and “not for production use.”

It is exclusively accessible in Google’s developer-facing tools, such as AI Studio.

Additionally, the model is not an ideal watermark eliminator. It seems that Gemini 2.0 Flash is having difficulty with semi-transparent watermarks and watermarks that cover significant areas of images.

Nevertheless, the absence of utilization restrictions in Gemini 2.0 Flash will undoubtedly be a source of contention for certain copyright holders.

Some models, such as Anthropic’s Claude 3.7 Sonnet and OpenAI’s GPT-4o, explicitly decline to remove watermarks. Claude characterizes the removal of a watermark from an image as “unethical and potentially illegal.”

According to law firms such as this one, the removal of a watermark without the original owner’s consent is illegal under U.S. copyright law, except a few rare exceptions.