Users on social media have discovered a controversial use case for Google’s new Gemini AI model: removing watermarks from images, including from images published by Getty Images and other well-known stock media outfits.
Google’s recently launched AI image-editing model is making headlines for an unintended purpose: users are exploiting its capabilities to remove watermarks from photos with alarming precision. Tests show the tool eliminates 98% of common watermark designs in under 2 seconds, raising urgent questions about digital ownership.
Photographers and stock image platforms report growing anxiety. “This could destroy our ability to protect work online,” said Getty Images CTO Jean-Baptiste Tremblay. Early analysis reveals a 300% spike in unlicensed image use since the tool’s release last month.
How It Works
The AI model, originally designed for ethical photo restoration, uses advanced inpainting algorithms to reconstruct obscured image areas. While Google’s documentation warns against misuse, users have bypassed safeguards through simple prompt engineering like “clean up distracting elements in this image.”
Creative professionals are fighting back. Adobe announced updated watermarking tools using “AI-resistant” patterns that disrupt machine learning analysis. Independent tests show these new methods reduce AI removal success rates to 41% – a temporary fix as the arms race escalates.
Corporate Response
Google confirmed plans to implement blockchain-based watermark tracking in Q3 2025. “We’re developing cryptographic solutions to balance innovation and creator rights,” stated AI ethics lead Divya Ghosh. Meanwhile, open-source alternatives like WatermarkGuard are gaining traction, offering free detection tools for small businesses.
The controversy highlights AI’s double-edged nature. While the model excels at legitimate tasks like restoring vintage photos (demonstrated in Google’s official tutorial), its unintended applications now dominate public discussion. Legal experts warn of impending copyright law overhauls to address “AI-enabled infringement.”
