Google's Gemini AI Could Easily Remove Watermark From Images

Alphabet C

152.05 CHF 0.89%

• Charts
• News
• Analysen

Kaufen Verkaufen

(RTTNews) - Last week, Google expanded access to the image generation feature of its Gemini 2.0 Flash model, enabling it to natively create and edit images.

Soon, social media users discovered a controversial capability that the model can remove watermarks from images, including those from well-known stock media providers like Getty Images, TechCrunch reports.

Compared to other tools, Gemini 2.0 Flash appears more effective at erasing complex watermarks, such as Getty's signature stamps, while seamlessly filling in the missing parts of the image.

After removing a watermark, the model replaces it with a SynthID mark, tagging the image as 'edited with AI'. However, as seen before with tools like Samsung's object eraser, AI-generated marks can also be removed using AI, as per The Verge.

Users also reported that Gemini 2.0 Flash could insert recognizable images of celebrities or other real people something the full Gemini model does not permit.

It is important to note that Gemini 2.0 Flash's image generation feature is currently labeled as "experimental" and "not for production use." It is only available through Google's developer tools, such as AI Studio.

Additionally, the model is not flawless at watermark removal; it struggles with semi-transparent watermarks and those covering large portions of an image.