In February 2023, a team of researchers at the University of Chicago released a program calledGlazea free software tool designed to use machine learning to fine-tune the pixels of user-supplied artwork, thereby altering its style to prevent any AI art-generating models from grabbing and training on it. The goal of this tool is to help artists prevent AI models from mimicking their art style without adversely affecting their own work. Not only do artists lose out on commissions and basic income as their style is intended to be copied, but low-quality synthesized copies distributed online can also weaken their brand and reputation.
The Glaze tool has gained traction, with over 740,000 downloads as of summer 2023. Now, more than a year later, the University of Chicago Glaze Project team has launched itsFirstThe new version of the product: Glaze2. Glaze2 is described as a faster and more protective tool, especially for new AI models such as Stable Diffusion XL.
The release of Glaze2 brings significant improvements, including increased image modification speeds from 50% to 500%, depending on the user's computer hardware. In addition, Glaze2 offers enhanced protection, especially for "smooth surface art" (e.g., anime, cartoons).
In addition, the Glaze project team plans to extend the tool's protection to cover short videos and animations. They are also conducting a user survey to evaluate AI's ability to replicate video frames and how well the new version of Glaze interferes with mimicry.
The release of Glaze2 marks an even greater step forward for artists in protecting their creations from unauthorized AI crawling, and will be an important tool for artists who want to protect their unique style.
Paper:https://arxiv.org/pdf/2302.04222.pdf