Google’s AI model used to erase watermarks from images, raising copyright concerns

Besides removing watermarks from existing photos, the model’s text-to-image generation feature also appears to have fewer guardrails and can reportedly be used to natively generate images depicting celebrities as well as other copyrighted content.

The purported lack of usage restrictions and removal of watermarks without consent could spur further legal action by copyright holders.

To be sure, access to Gemini 2.0 Flash’s image generation feature was expanded last week for “experimental” purposes  and “not for production use”. Currently, it can only be accessed via Google AI studio, the tech giant’s platform for hosting AI developer tools.

In 2024, Google DeepMind announced that it was making SynthID Text, its tool used to watermark AI-generated text, publicly available. Last month, it announced that its Photos app (Google Photos) will begin using SynthID to mark AI-manipulated images.

The Alphabet-owned company is also part of the Coalition for Content Provenance and Authenticity (C2PA) along with Amazon, Microsoft, OpenAI, Adobe, and others.

The C2PA group has developed technical standards to help trace the history of AI-generated images by providing metadata with details such as date of creation and AI tools used to create the content. However, it faces challenges of adoption and interoperability.