Social media users have identified a controversial use case for Google’s Gemini AI model, specifically its ability to remove watermarks from images, including those from Getty Images and other well-established stock media companies.
Google recently expanded the capabilities of its Gemini 2.0 Flash model, allowing broader access to its image generation feature. This feature enables the model to create and edit images natively. While its capabilities are deemed powerful, they appear to include minimal safeguards. Gemini 2.0 Flash is capable of generating images depicting celebrities and copyrighted characters and can effectively remove watermarks from existing photos.
Various users on platforms such as X and Reddit have noted that Gemini 2.0 Flash not only removes watermarks but also attempts to seamlessly fill any voids left by their deletion. Although other AI tools can perform this task, Gemini 2.0 Flash stands out for its proficiency and availability at no cost to users.
While Gemini 2.0 Flash is termed “experimental” and “not for production use,” and is accessible only through Google’s developer-focused tools like AI Studio, its function as a watermark remover is not infallible. The model faces challenges with semi-transparent watermarks and those that cover extensive areas of an image.
Despite these limitations, some copyright holders are likely to express concerns over the unrestricted nature of Gemini 2.0 Flash’s usage. Other AI models, such as Claude 3.7 Sonnet by Anthropic and OpenAI’s GPT-4o, explicitly refuse to remove watermarks, with Claude describing the act as “unethical and potentially illegal.”
Under U.S. copyright law, removing a watermark without the consent of the original owner is generally considered illegal, a stance supported by various legal sources.
Google has not provided an immediate response to requests for comment submitted outside of standard business hours. However, as of March 17, a Google spokesperson stated: “Using Google’s generative AI tools to engage in copyright infringement is a violation of our terms of service. As with all experimental releases, we’re monitoring closely and listening for developer feedback.”
