Users on social media have found a controversial use case for Google’s new Gemini AI mannequin: eradicating watermarks from pictures, together with from pictures revealed by Getty Images and different well-known inventory media outfits.
Last week, Google expanded entry to its Gemini 2.0 Flash mannequin’s picture era function, which lets the mannequin natively generate and edit picture content material. It’s a robust functionality, by all accounts. But it additionally seems to have few guardrails. Gemini 2.0 Flash will uncomplainingly create pictures depicting celebrities and copyrighted characters, and — as alluded to earlier — take away watermarks from current images.
New talent unlocked: Gemini 2 Flash mannequin is absolutely superior at eradicating watermarks in pictures! pic.twitter.com/6QIk0FlfCv
— Deedy (@deedydas) March 15, 2025
As a number of X and Reddit customers famous, Gemini 2.0 Flash received’t simply take away watermarks, however can even try to fill in any gaps created by a watermark’s deletion. Other AI-powered instruments do that, too, however Gemini 2.0 Flash appears to be exceptionally expert at it — and free to make use of.
Gemini 2.0 Flash, accessible in Google’s AI studio, is superb at modifying pictures with easy textual content prompts.
It can also take away watermarks from pictures (and places its personal delicate watermark in as a substitute 🤣) pic.twitter.com/ZnHTQJsT1Z
— Tanay Jaipuria (@tanayj) March 16, 2025
To be clear, Gemini 2.0 Flash’s picture era function is labeled as “experimental” and “not for manufacturing use” in the intervening time, and is simply accessible in Google’s developer-facing instruments like AI Studio. The mannequin additionally isn’t an ideal watermark remover. Gemini 2.0 Flash seems to battle with sure semi-transparent watermarks and watermarks that canvas giant parts of pictures.
Still, some copyright holders will certainly take challenge with Gemini 2.0 Flash’s lack of utilization restrictions. Some fashions, together with Anthropic’s Claude 3.7 Sonnet and OpenAI’s GPT-4o, explicitly refuse to take away watermarks; Claude calls eradicating a watermark from a picture “unethical and probably unlawful.”
Removing a watermark with out the unique proprietor’s consent is taken into account unlawful below U.S. copyright legislation (in response to legislation corporations like this one) exterior of uncommon exceptions.
Google didn’t instantly reply to a request for remark despatched exterior of regular enterprise hours.
Updated 3/17 at 1:48 p.m. Pacific: A Google spokesperson supplied the next assertion:
“Using Google’s generative AI instruments to interact in copyright infringement is a violation of our phrases of service. As with all experimental releases, we’re monitoring carefully and listening for developer suggestions.”
