Google's Gemini 2.0 Flash is reportedly capable of removing watermarks — Also found to generate AI celebrity photos

When you buy through links on our articles, Future and its syndication partners may earn a commission.

Credit: Google

Google's Gemini 2.0 Flash model can reportedly remove watermarks from user-provided images and depict copyrighted characters/celebrities in generated images, claim users on Reddit (via ComputerBase). This includes licensed stock imagery found online, stamped with watermarks to indicate their copyrighted nature. This risks potential litigation from media licensing corporations. Limited testing on our end revealed that watermark removal was unsuccessful on several pr ovided images, while the generation of celebrity images was possible.

Back in December, Google introduced native image output with Gemini 2.0 Flash for testing. Just last week, this update was made accessible across all regions, free of charge, and includes a slew of new features such as: telling a story with pictures, conversational image editing, world understanding, text rendering, and more. AI image generators have consistently faced scrutiny regarding ethical and moral considerations.

Companies like Google, Anthropic, and OpenAI typically integrate a set of guardrails or a list of dos and don'ts in their models to combat the misuse of their technologies. The restrictions stretch wide, but the general prohibitions typically include depictions of celebrities and notably the removal of watermarks from copyrighted material.

Recently, several Redditors found that Gemini's native image output feature extends to watermark removal from stock images, such as those from Getty Images, and celebrity image generation as well. In the first case, this probably violates multiple DMCA provisions as removing a watermark alters the CMI, which is information that identifies copyrighted work. The legality of generating celebrity images using AI is slightly more complex, primarily concerned over the protection of celebrities' rights of publicity and concerns surrounding deepfakes and potential defamation.

Even so, the removal of the watermark doesn't change the fact the image is likely protected by copyright law. So you can't expect to use these images commercially, as you'd probably be hit with a barrage of DMCA claims. In most cases, individuals may desire to remove the watermark, if said image is intended for personal use-cases.

Contrary to the Reddit thread, we were not able to remove watermarks from the same images, with Gemini responding, "Unfortunately, I cannot directly edit images." Conversational image editing works as intended in Google AI Studio, where Gemini managed to reproduce the image of a celebrity, so there's still some ironing out left to be done.

It's important to note that Google has tagged the image generation feature of Gemini 2.0 Flash as experimental. This may be why it's only available in Google AI Studio and not on Gemini's main site and could also explain these lapses in judgment.