A new controversy surrounding the boundaries of AI is making waves in the tech world. Google's newly launched AI model, Gemini 2.0 Flash, has been found to possess a concerning ability: it can effortlessly remove image watermarks, including professional watermarks from renowned stock photo agencies like Getty Images. This functionality has sparked widespread concern among creators and copyright holders.
Last week, Google expanded access to Gemini 2.0 Flash's image generation capabilities, allowing more users to experience its powerful image generation and editing features. However, this decision seems to lack necessary safeguards. Netizens quickly discovered that Gemini 2.0 Flash not only generates images containing celebrities and copyrighted characters but, more alarmingly, can easily remove watermarks from existing photos.
Multiple users on X and Reddit confirmed that Gemini 2.0 Flash intelligently attempts to fill in the gaps left by watermark removal. While other AI-based watermark removal tools exist, Gemini 2.0 Flash excels in this area and is currently completely free for users, significantly increasing potential copyright risks.
It's noteworthy that Google labels Gemini 2.0 Flash's image generation capabilities as "experimental" and "non-production use," only offering it within developer-focused tools like AI Studio. Furthermore, the model isn't a perfect watermark remover, particularly struggling with semi-transparent watermarks or those covering large areas of the image.
Despite this, copyright holders express strong concerns about this feature's lack of usage restrictions. In contrast, competing models like Anthropic's Claude 3.7 Sonnet and OpenAI's GPT-4 refuse watermark removal. Claude even explicitly labels watermark removal as "unethical and potentially illegal." Indeed, under US copyright law, removing watermarks without the original owner's consent is generally illegal, except in rare circumstances.
As of now, Google remains silent amidst this controversy. This incident again prompts deep reflection on the boundaries of AI technology, the balance between commercial interests, and the rights of creators.
