Nightshade, a tool that prevents AI art generators from plagiarizing publicly viewable content, became available on Jan. 18 after several months of development and peer review. Since then, it’s become clear that artists, photographers, and others sharing images online are eager to protect their work from generative AI. The team behind Nightshade reports that in the five days following its release, Nightshade received a quarter of a million downloads.
“I expected…extremely high enthusiasm,” Ben Zhao, Nightshade’s project lead and a University of Chicago computer science professor, told VentureBeat. “But I still underestimated it…The response is simply beyond anything we imagined. Based on reactions on social media, the downloads come from all over the globe.”
At this point, any art or photos you post on the internet will get sucked into a generative AI training dataset—even if you’ve copyrighted your work. The only way to prevent AI art generators from ripping off your content is to run your images through a data-poisoning application. That’s where Nightshade comes in. After you upload an image to Nightshade, the tool manipulates the data built into that image, making it unsuitable for model training. To the human eye, most “shaded” images look the same as they did pre-Nightshade; only art without much texture might reveal a few odd pixels here and there if the artist has neglected to use Nightshade’s “low intensity” setting.
Shading an image prevents the data scrapers that serve AI models from being able to interpret the image’s contents correctly. On their website, Nightshade’s creators provide the example of a cow in a green field, which might look to AI like a leather purse lying in the grass. An image of a dog becomes an image of a cat, and so on. Even if a shaded image has been cropped, compressed, or visually altered with an application like Photoshop, generative AI tools like DALL-E and Stable Diffusion cannot tell what the original image contained.
It’s important to note that for some artists, Nightshade alone might not be enough to keep AI’s grubby hands off their work. While Nightshade alters the apparent content of an image behind the scenes, it takes another tool to prevent AI from mimicking an artist’s style. Glaze, Nightshade’s older sibling and the original tool from the University of Chicago team, works similarly to Nightshade in that it manipulates an image’s data without changing much of its appearance. But instead of making it appear as though a work of art depicts a different subject, it changes its visual style. A charcoal portrait, for example, might look to AI like a modern abstract Jackson Pollock copycat. Therefore, using both Nightshade and Glaze does more to ensure that an artist’s work will remain useless to AI data scrapers.
The Nightshade team also says they’re less focused on ruining AI art generators than they are on recentering artists. “Nightshade's goal is not to break models, but to increase the cost of training on unlicensed data, such that licensing images from their creators becomes a viable alternative,” their website reads.
Both Nightshade and Glaze are available for free on the team’s website. They’re also available via this X post, which the developers made after Nightshade’s copious downloads saturated their campus network link. (A note of caution: Clicking either of the links in the post will initiate an instant download.) Within the coming months, the team plans on creating a tool that combines Nightshade’s and Glaze’s functionalities, making them easier to use simultaneously.
