Adapted by:
Javier Pastor
Senior WriterComputer scientist turned tech journalist. I've written about almost everything related to technology, but I specialize in hardware, operating systems and cryptocurrencies. I like writing about tech so much that I do it both for Xataka and Incognitosis, my personal blog.
Alba Mora
WriterAn established tech journalist, I entered the world of consumer tech by chance in 2018. In my writing and translating career, I've also covered a diverse range of topics, including entertainment, travel, science, and the economy.
Using watermarks is a common practice when someone wants to protect their images for selling or distributing them securely. However, Google’s new AI model might compromise this protection. Gemini 2.0 Flash has shown the ability to remove watermarks in certain instances. Naturally, the implications of this are significant.
Generating images with Gemini 2.0 Flash. Google’s Gemini 2.0 Flash family of AI models has been making headlines since its debut. It introduced a preliminary reasoning mode called Flash Thinking. More recently, it debuted an intriguing image generation mode.
Gemini’s image generator presents some striking options along with a controversial feature. Users can access it in Google AI Studio by selecting this mode in the Create Prompt section. You’ll find it by scrolling down the Model dropdown menu on the right side. Simply choose Gemini 2.0 Flash (Image Generation) Experimental.
Multimodal. One of the most notable features of Gemini’s image generator is its multimodal capability. Traditionally, generating an image requires a text prompt to describe the desired outcome. However, with Gemini 2.0 Flash and this new mode, users can generate images based on other images. For example, if there’s a picture of someone wearing a white T-shirt, you could modify it by requesting, “Make the T-shirt red.”
Surprising but imperfect. In my tests, I uploaded the image above and modified the girl’s hair and jacket. While the tool successfully changed the hair to blonde as requested, it resulted in a complete alteration of the girl’s features. I then asked the AI model to change the jacket to a blue T-shirt. The outcome wasn’t perfect either, but the final result was still striking.
Click on the image to view the original post on X.
Removing watermarks. A much more surprising yet controversial capability of Google’s new model is its ability to remove watermarks from images. Many users have shared their experiments on X, demonstrating how in some cases the watermarks have completely disappeared.
In this example, the AI model not only failed to remove the watermarks, but it also completely changed the model’s face.
Not 100% effective. To test the watermark removal feature, I added a watermark to an image and asked Gemini 2.0 Flash to remove it. As shown in the image above, it failed to remove the watermark. It also altered the girl’s face and removed some elements, like the small pink ice cream cone in the bottom right corner. In this example, the generative AI model seems to make its own unpredictable and often incorrect decisions.
When it works, it’s mind-blowing. In further tests, the results were often surprising. In one case, I took a watermarked image, cropped it, and tested the model. As seen in the image below, the result was impressive, but there were some drawbacks. Items like the girl’s bracelets and pendant disappeared. Despite this, the method proved effective, raising concerns about its implications.
AI and copyright. OpenAI recently requested that the U.S. government lift copyright restrictions for AI companies. This move would allow for the unrestricted use of copyrighted works without consequences. Adding to the controversy, Google’s generative AI model has features that may unintentionally facilitate the removal of watermarks. It’s concerning, especially for major image banks that rely on these markings to ensure artists can profit from their creations.
Google’s response. In response to these concerns, a Google spokesperson told Xataka: “Using Google’s generative AI tools for copyright infringement is a violation of our terms of service. As with all experimental releases, we’re closely monitoring the situation and paying attention to developer feedback.”
Image | Google
Log in to leave a comment