Social media users have discovered a controversial use for Google's new Gemini artificial intelligence model: removing watermarks from images, including those published on Getty Images and other well-known stock media resources.
Social media users have discovered a controversial use for Google's new Gemini artificial intelligence model: removing watermarks from images, including those published on Getty Images and other well-known stock media resources.
Last week, Google expanded access to the image generation feature for its Gemini 2.0 Flash model, which allows the model to generate and edit image content on its own. By all accounts, this is a powerful feature. But it also seems to have a few drawbacks. Gemini 2.0 Flash creates images from celebrity and copyrighted photos without any complaints, and also removes watermarks from photos, writes TechCrunch.
New skill unlocked: Gemini 2 Flash model is really awesome at removing watermarks in images! pic.twitter.com/6QIk0FlfCv
— Deedy (@deedydas) March 15, 2025
As several X and Reddit users have pointed out, Gemini 2.0 Flash doesn't just remove watermarks, it also attempts to fill in any gaps left after the removal. Other AI tools do this as well, but Gemini 2.0 Flash seems to be exceptionally good at it — and it's free to use.
Gemini 2.0 Flash, available in Google's AI studio, is amazing at editing images with simple text prompts.
— Tanay Jaipuria (@tanayj) March 16, 2025
It also can remove watermarks from images (and puts its own subtle watermark in instead 🤣) pic.twitter.com/ZnHTQJsT1Z
The image generation feature in Gemini 2.0 Flash is currently marked as "experimental" and "not for production use," and is only available in Google's developer tools, such as AI Studio. The model also doesn't perfectly remove watermarks. Gemini 2.0 Flash seems to have trouble handling certain semi-transparent watermarks and watermarks that overlap large parts of images.
However, some copyright holders will likely be unhappy with the lack of restrictions on the use of Gemini 2.0 Flash. Models such as Anthropic’s Claude 3.7 Sonnet and OpenAI’s GPT-4o explicitly refuse to remove watermarks; Claude calls removing a watermark from an image “unethical and potentially illegal.”
