MANILA, Philippines – Generative AI image tools have the bad habit of scraping – that is essentially mimicking or copying – artist styles for use in the tools’ created images.

While artists spend years honing their craft to create styles that can be easily seen as their own, an AI tool can mimic that so long as it has the data that allows it to categorize an artist’s style for its own use.

A new application, Glaze, aims to help protect artists’ styles from being scraped by AI image generators like Midjourney or Stable Diffusion, by “cloaking” the image to make it more difficult for AI to scrape it.

Glaze was developed as part of a research project (under peer review and readable here) by PhD students and computer science professors at the University of Chicago who wanted to protect internet users “from invasive uses of machine learning” and “explore how ethical security techniques can be utilized to develop practical solutions and (hopefully) help real users.”

In a TechCrunch report, Ben Zhao, a professor of computer science at University of Chicago, who is the faculty lead on the project, explained how Glaze works.

Zhao explained, “What we do is we try to understand how the AI model perceives its own version of what artistic style is. And then we basically work in that dimension – to distort what the model sees as a particular style. So it’s not so much that there’s a hidden message or blocking of anything… It is, basically, learning how to speak the language of the machine learning model, and using its own language – distorting what it sees of the art images in such a way that it actually has a minimal impact on how humans see. And it turns out because these two worlds are so different, we can actually achieve both significant distortion in the machine learning perspective, with minimal distortion in the visual perspective that we have as humans.”

Zhao added that the gap between how humans and AI models see the world, “using mathematical representation, seems to be fundamental and unavoidable.” 

“What we’re actually doing – in pure technical terms – is an attack, not a defense. But we’re using it as a defense,” he said.

Glaze’s limits

Glaze’s ability to defend style theft, while admirable, is also limited and can quite possibly end up being bit of a development arms race.

The app’s developers mention in a list of limitations that Glaze’s changes to a person’s created art “are more visible on art with flat colors and smooth backgrounds, e.g. animation styles. While this is not unexpected, we are searching for methods to reduce the visual impact for these styles.” They also note that due to the pace of AI development, the art the tool can protect now may not be protected later on because of advancements in scraping that fight against the cloaking mechanism done by Glaze.

Despite the limitations, the developers have said that, “to the best of our knowledge, this is the only available tool for artists to proactively protect their style while posting their art online today. Note that this is ongoing research, and we will continuously update our tools to improve its robustness.” – Rappler.com