Gemini’s image generator seems to excel at removing watermarks

Recently, Google expanded access to Gemini 2.0 Flash’s image generation capabilities. The model fully utilizes Google’s latest image synthesis tool, Imagen 3, to give users the ability to generate and edit images. Now that users have had some time to test out the feature, they’re noticing it can do some things it probably shouldn’t be able to do.

It didn’t take long for people to begin testing the boundaries of Gemini 2.0 Flash’s image-generation feature. While the tool is capable of generating some impressive results, it also appears to be lacking some important guardrails. Users are finding that the model has zero qualms about depicting celebrities or copyrighted characters. It also appears to be quite skilled at removing watermarks from images.

Several users on Reddit and X (formerly Twitter) have pointed out Gemini’s ability to remove watermarks (h/t: News). It not only gets rid of the watermark, but also does a pretty good job of replacing what was removed after the watermark is deleted. However, it appears to struggle a bit with semi-transparent watermarks and large watermarks that take up a majority of the image.

We have to imagine that this functionality is not being well-received by copyright holders. It’s important to keep in mind that watermarks usually serve as a visual representation that the image is someone else’s property. As such, removing a watermark could be considered copyright infringement. This is why AI models like Anthropic’s Claude have guardrails that prevent such action.

Gemini 2.0 Flash’s image generation feature is currently only available in developer tools like AI Studio. It also carries the labels “experimental” and “not for production use.” It wouldn’t be too surprising if Google makes some quick changes to the feature to prevent watermark removal.