Social media users have found a controversial function in Google’s Gemini AI model—its ability to erase watermarks from images, including those from major stock image providers like Getty Images.
Recently, Google introduced expanded access to the image generation feature of its Gemini 2.0 Flash model, enabling users to create and modify images with ease. While the feature is powerful, it lacks sufficient safeguards. Gemini 2.0 Flash can generate images featuring celebrities or copyrighted characters and notably has the capability to remove watermarks from existing images.
Users on platforms like X and Reddit have reported that Gemini 2.0 Flash not only eliminates watermarks but also cleverly fills in the space where the watermark once appeared. While other AI tools can perform similar tasks, Gemini 2.0 Flash is said to execute the task exceptionally well, and it is available for free.
It’s worth mentioning that the image generation function of Gemini 2.0 Flash is currently labeled as “experimental” and “not for production use.” It can only be accessed through Google’s developer-focused tools, like AI Studio. Although the model is impressive, it does have some limitations, such as difficulty handling semi-transparent watermarks or those that cover a large portion of an image.
The lack of restrictions surrounding the Gemini 2.0 Flash model could raise significant concerns for copyright holders. In contrast, other models, such as Anthropic’s Claude 3.7 Sonnet and OpenAI’s GPT-4o, explicitly refuse to remove watermarks, with Claude calling it “unethical and potentially illegal.”
Under U.S. copyright law, watermark removal is considered illegal without the permission of the original owner, with only a few rare exceptions. Most AI companies that generate images add watermarks to indicate their source. However, an unexpected consequence has emerged with these AI tools: they can now remove watermarks from stock image databases, a task Gemini 2.0 Flash appears to handle particularly well.
Users have shared examples of watermark removal using Gemini 2.0 Flash on Reddit and X, showing how the AI effortlessly removes various watermarks, seamlessly blending the images. The AI-generated images include a watermark at the bottom left, indicating the nature of AI-created content. However, TechCrunch has reported that this feature is still marked as “experimental” and available only through tools like AI Studio.
The lack of regulations on Google’s Gemini 2.0 Flash model extends to its ability to generate images of celebrities and copyrighted characters with minimal restrictions. This is particularly concerning given past incidents in which AI-generated content led to controversies involving renowned artists.
Featured image credit: Mitchell Luo/Unsplash
