Google’s AI is being used to remove watermarks from stock photo images

Gemini

Some people always find a way to bend technology to their will? Well, the internet has done it again—this time with Google’s latest AI model, Gemini 2.0 Flash.

Originally designed to generate and edit images, Gemini 2.0 Flash has a special trick–it can remove watermarks from images. And yes, that includes those from big stock photo sites like Getty Images. People on X (formerly Twitter) and Reddit were quick to notice and test it out. Turns out, Gemini doesn’t just erase watermarks, it also magically fills in the missing parts of the image, making it look like the watermark was never there in the first place.

Now, before you start celebrating free stock images, there’s a catch. Removing a watermark without permission is illegal in most cases—US copyright law is pretty clear on that. Other AI models, like OpenAI’s GPT-4o and Anthropic’s Claude 3.7, straight-up refuse to do it, calling it “unethical.” But Gemini, for now, is successfully removing watermarks from pictures.

Google does label Gemini’s image generation feature as “experimental” and “not for production use.” Plus, it’s currently only available through Google’s AI Studio, a tool meant for developers. The model isn’t flawless at watermark removal—it struggles with semi-transparent ones or those covering large parts of an image. But the fact that it works at all is enough to raise eyebrows.

The copyright holders aren’t thrilled. Big companies spend a lot of money protecting their images, and an AI that casually erases their watermarks isn’t exactly good news for them. So, will Google step in and tighten the rules? Maybe. But for now, the internet is doing what it does best—pushing AI to its limits, one controversial experiment at a time.Google has not commented yet.