Google’s Latest Gemini Image Editor Can Remove Watermarks From Photos
Rapid progress in artificial intelligence continues to dominate the headlines, but the latest story to hit the news cycle is more bad than good: It turns out that Google’s latest Gemini 2.0 Flash image editor is pretty good at automatically removing watermarks from images.
A few days ago, Google introduced experimental native image generation capabilities, pushing its capabilities in combining images and text, editing images with conversational prompts, and improving its “understanding of the world” to provide users with AI-generated images that are overall more realistic.
As TechCrunch reports, the newly updated tool was quickly put to use for one specific and rather nefarious use: removing watermarks from proprietary images. You can find evidence of this on Reddit and X ; While removing the watermark isn’t perfect—after all, the AI is only imagining the pixels that replace what’s covered by the watermark—it definitely leaves you with a usable image, free of any copyright or attribution marks.
It’s not entirely ironic that an artificial intelligence tool would be so capable of circumventing basic copyright protection. Gemini and other generative AI models were trained on massive amounts of copyrighted text, images and video, often without permission or compensation—something AI companies are reluctant to talk about unless they claim it qualifies as fair use .
This tweet is currently unavailable. It may be downloading or has been deleted.
While the idea of using artificial intelligence trained on copied copyrighted works to steal someone else’s image is annoying, there are a few caveats worth mentioning. For starters, these new Gemini tools are only available to developers for now and are still labeled as experimental—no doubt Google is going to make changes before regular users get their hands on them. It’s also worth noting that other dubious watermark removal tools are already out there all over the web, even if none of them are as smart as this and other advanced AI-powered versions.
However, it highlights the nefarious ways in which AI tools are being used, even as they are relentlessly promoted to businesses and consumers, and highlights the need for strong protections. The ChatGPT and Claude AI models are two models that will refuse explicit requests to remove watermarks, and no doubt Google will add the same blocks to Gemini after all the negative coverage.
Limitations of AI Image Editing
As noted above, Gemini’s watermark removal capabilities can currently only be accessed through developer tools, including AI Studio and the Gemini API. However, I also wanted to try using the latest models available through the Gemini Advanced tools, available to anyone who pays $20 per month .
I took screenshots of a few copyrighted and watermarked images from Shutterstock (after previewing all the AI-generated results the site offered me), opened and resave them in Photoshop (after closing all the pop-ups asking me to try Photoshop’s latest AI tools), and then let Gemini get to work.
Most of the time, Gemini refused to engage in watermark removal at all, telling me it couldn’t edit images or explaining that “copyright laws must be followed” – but it was happy to create variations of the originals, without watermarks and based on its training on copyrighted material .
While these versions often weren’t all that similar to the originals, in one case they gave me a very accurate copy of a composite photo of a rocket ship, and in another they directed me to some watermark removal tools on the Internet—thanks, Gemini! Of course, you can also start from scratch and simply describe the watermarked image you hope to imitate.
Meanwhile, ChatGPT accepted requests to remove watermarks, but then returned something completely different (and quite strange). Image editing is already available to ChatGPT Plus subscribers, but as I noted earlier , it’s definitely not at the level of a Photoshop expert yet.
The debate over copyright, fair use, and AI safety measures will continue, even as update after update makes these AI tools more advanced (even if they aren’t particularly useful ). One issue that companies like Google and OpenAI may need to start worrying about is where they’ll get fresh training data for their models once they push out all the flesh-and-blood creatives.