A controversial application of Google’s new Gemini AI model has been uncovered by social media users, who have found that it can be used to remove watermarks from images, including those from prominent stock media companies like Getty Images.
Recently, Google granted broader access to the image generation feature of its Gemini 2.0 Flash model, enabling the model to natively generate and edit image content. This capability is undoubtedly powerful, but it appears to lack sufficient safeguards. The Gemini 2.0 Flash model can create images depicting celebrities and copyrighted characters without complaint, and it is also capable of removing watermarks from existing photos.
A new skill has been discovered: the Gemini 2 Flash model is highly effective at removing watermarks from images! pic.twitter.com/6QIk0FlfCv
— Deedy (@deedydas) March 15, 2025
Several users on X and Reddit have noted that the Gemini 2.0 Flash model not only removes watermarks but also attempts to fill in any gaps created by the deletion process. While other AI-powered tools have this capability, the Gemini 2.0 Flash model seems to be exceptionally skilled at it, and it is free to use.
The Gemini 2.0 Flash model, available in Google’s AI studio, is amazing at editing images using simple text prompts.
It can also remove watermarks from images and replace them with its own subtle watermark 🤣 pic.twitter.com/ZnHTQJsT1Z
— Tanay Jaipuria (@tanayj) March 16, 2025
It is worth noting that the image generation feature of the Gemini 2.0 Flash model is currently labeled as “experimental” and “not for production use,” and it is only available in Google’s developer-facing tools like AI Studio. Additionally, the model is not perfect and struggles with certain semi-transparent watermarks and watermarks that cover large portions of images.
Despite this, the lack of usage restrictions in the Gemini 2.0 Flash model may be a concern for some copyright holders. Some models, such as Anthropic’s Claude 3.7 Sonnet and OpenAI’s GPT-4o, explicitly refuse to remove watermarks, with Claude considering it “unethical and potentially illegal.”
Removing a watermark without the consent of the original owner is considered a violation of U.S. copyright law, with few exceptions, according to law firms and legal experts.
Google did not immediately respond to a request for comment sent outside of normal business hours.
Updated 3/17 at 1:48 p.m. Pacific: A Google spokesperson provided the following statement:
“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 are closely monitoring and listening to developer feedback.”
