The Pixel 9 series launched last August came with impressive Gemini AI features, including image generation abilities that were too good, opening the door to abuse. It was troubling to see Google need about two months to start fixing the Pixel Studio and Reimagine issues.
Those fixes, labeling Gemini-made images as images created with AI, should have been in place from the start, but they’re still not enough to prevent abuse.
Fast-forward to mid-March and Google has given Gemini a big upgrade, making it more personal and improving Deep Research. Gemini 2.0 Flash Thinking Experimental was the model Google highlighted in its announcement, the reasoning AI that makes those research reports better and helps with Gemini personalization.
It turns out that Gemini 2.0 Flash is also very good at creating and editing images. Too good. When I say that, I mean it can be abused with ease.
Specifically, Gemini 2.0 Flash users found they could remove watermarks from images on the web with incredible ease. This essentially allows any user to steal copyright-protected imagery from the web. It’s a lawsuit waiting to happen.
You might have seen Gemini 2.0 Flash images on the web via social media posts meant to highlight the impressive abilities of Google’s image-generation tools. You’ll find the Gemini 2.0 Flash image generation tool in Google’s AI Studio, which is free to use.
Some Gemini 2.0 Flash users found that you can tell the AI to remove the watermarks in photos, and the AI will comply. The results are impressive, even with photos featuring plenty of watermarks.
If you’ve ever bought copyright images on the web, you know they’re usually covered in plenty of watermarks to prevent users from cropping parts of them. Gemini 2.0 Flash removes those watermarks in a, well, flash, as you can see in the images above.
The results aren’t perfect, but they’re good enough. Google probably never intended for users to abuse Gemini 2.0 Flash this way and remove watermarks. But Gemini does its job remarkably well.
Ironically, the AI marks the newly generated result with a Gemini watermark in a corner to inform people the image was created with AI. That’s a good safety measure, though one could easily crop that corner watermark out.
I will point out what others said on Reddit: You can remove watermarks from photos with other software. It’s likely that other AI programs exist to do just that.
However, the Gemini 2.0 Flash’s ability to remove watermarks makes the process trivial. You don’t have to teach yourself new skills to remove watermarks from an image protected by copyright. You only have to give Gemini a written command.
While I haven’t replicated the Gemini 2.0 Flash watermark removal feature myself, others have been able to use it. I think it’s definitely a bug that Google will fix in the not-too-distant future. I wouldn’t be surprised if future versions of AI Studio will not remove watermarks.
I’m surprised Google didn’t consider this problem before it released the Gemini 2.0 Flash upgrades, especially after all the backlash regarding the Pixel 9’s image-generation abilities that lacked safeguards. Then again, giving the AI this sort of freedom will certainly generate free publicity that might be warranted in the incoming lawsuits.
