Google’s newest AI model, Gemini 2.0 Flash, is exceptionally good at removing watermarks from copyrighted images including Getty Images, according to multiple users. However, AI watermark removal does raise copyright concerns.
Gemini 2.0 Flash is a developer-facing AI tool
Gemini 2.0 Flash is free to use, which further increases the AI model’s appeal; it is only available through Google’s developer-facing AI tools, such as Google AI Studio. To alter a preexisting image with the Gemini model, or to create a new image, users must select “Gemini 2.0 Flash (Image Generation) Experimental.” Users must also ensure the “Images and text” output form is selected, rather than just text.
Gemini 2.0 Flash fills in removed watermarks with smart AI
The Gemini 2.0 Flash model doesn’t just eliminate the watermark — it also fills in that space to blend the watermark’s removal. All generative AI tools attempt this process to some extent, but Gemini appears to be more advanced in filling in gaps seamlessly. Gemini does add its own subtle watermark to the final image: a small blue diamond in the bottom left hand corner.
In addition, the AI model sometimes struggles to completely remove transparent and/or very large watermarks.
AI watermark removal raises copyright concerns
The Gemini 2.0 Flash model is labeled “experimental” and “not for production use.” The AI model’s abilities suggest it currently has few guardrails to limit its images creation and modification abilities; this is especially notable since several other generative AI models, such as Anthropic’s Claude 3.7 Sonnet and OpenAI’s GPT-4o, do not allow watermark removal.
Removing a watermark without permission from the original image owner is considered illegal under U.S. copyright laws, except in rare cases. When asked to comment on the situation by TechCrunch, 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’re monitoring closely and listening for developer feedback.”