Social media users have discovered a controversial capability of Google’s new generative AI model, Gemini 2.0 Flash — removing watermarks from images, including those published by well-known companies such as Getty Images. Last week, Google expanded access to this model’s image generation feature, allowing users to create and edit image content. However, it turned out that the model has minimal usage restrictions. Gemini 2.0 Flash can not only generate images featuring celebrities and copyright-protected characters, but also remove watermarks from existing photos.
Several users noted that Gemini 2.0 Flash doesn’t just remove watermarks, but also tries to fill in the gaps left by their removal. While other AI tools can also perform this function, Gemini 2.0 Flash has proven to be particularly adept at it. The model is currently available for free in Google’s AI Studio, where users can edit images using simple text prompts.
It’s worth noting that the image generation feature of Gemini 2.0 Flash is labeled as “experimental” and “not for production use” and is only available through developer tools. The model doesn’t always remove watermarks perfectly — it struggles with semi-transparent watermarks or those covering large portions of an image. Nevertheless, the lack of restrictions on this feature’s use may raise concerns among rights holders.
A Google spokesperson stated that using generative AI tools to infringe copyright is a violation of the company’s terms of service. Google is closely monitoring experimental releases and taking developer feedback into account. Other models, such as Claude 3.7 Sonnet by Anthropic and GPT-4o by OpenAI, explicitly refuse to remove watermarks, considering it unethical and potentially illegal.
