Students from the University of Chicago have developed software that allows the generation of a “masked” version of an original work (drawings, paintings, etc.), to protect it from artificial intelligence (AI) programs, a new tool at the service of artists
outraged artists
Called Glaze, the solution, available for free on March 15, adds a layer of watermark information nearly invisible to the naked eye. The author can then post the modified image online. If a generative AI model tries to use it, the added computational data will prevent the machine from correctly parsing the style and playing it.
The SAND Lab at the University of Chicago decided to design Glaze after being contacted by artists outraged that programs like Stable Diffusion or Midjourney allow their users to generate images in seconds that imitate their style in a very sophisticated way.
“We work on the security and privacy issues of AI systems, but normally we work more on tools ‘in case such and such a problem arises,'” he told AFP. “This time the problem already exists, it causes real damage to artists, they need tools to protect them.”
Identify art differently
The team of researchers and engineers worked in collaboration with artists, including Karla Ortiz, one of the plaintiffs in the San Francisco class action lawsuits against several companies.
“If someone then asks Stable Diffusion to produce ‘a Karla Ortiz-esque piece,’ they’ll get van Gogh-esque images (or something hybrid).”
The team acknowledges, however, that Glaze is not a permanent solution against AI mimicking. Specialized companies will certainly be able to get around the obstacle, forcing students to improve their tool, and so on.
“AI moves too fast, too fast,” says Shawn Shan. “The idea is to buy time while waiting for more regulation.”
Source: BFM TV
