The free tool Glaze, designed to protect artists from having their style copied by AI image generators, is currently experiencing a dramatic surge in demand.
The tool adds imperceptible noise to images to prevent AI systems from imitating the style. According to Ben Zhao, the developer of Glaze, there is a large backlog of access requests for the web-based version WebGlaze, since Meta announced plans to use user data for AI training.
Artists sometimes have to wait weeks or months for access - the Glaze Project manually reviews each application to ensure they are real people and the tools are not being abused. At the same time, security researchers have found a way to bypass the protection of Glaze.
Although Zhao and his team have made changes to make the attack more difficult, the attack still calls into question the effectiveness of Glaze - especially since the team behind the attack has criticized the changes as insufficient.
Glaze is a free tool created to protect artists from having AI image generators copy their artistic style. The tool works by adding invisible noise to images, which tricks AI systems into perceiving a different style than the original. This prevents the AI from imitating the artist's unique style.
The team behind Glaze is also developing another tool called Nightshade. This tool aims to corrupt AI models by using manipulated pixels to confuse them. For instance, an AI model might identify a train as a car due to these altered pixels. Fewer than 100 of these "poisoned" images could potentially corrupt an AI image model or its training data.
The development of these tools is a response to the ongoing debate surrounding AI and copyright. Many large AI companies have used images freely available online to train their image AI models without obtaining permission from the creators. Artists see this as a potential violation of their copyrights and feel that their work is being threatened by these practices.
