By Johnathan Subramanian
Over the past few years, the popularity of AI image generation has grown exponentially. This growth has been met with great concern from human artists regarding the ethics and copyright protections related to these AI art services. In response to these concerns, new technologies have been invented to help human artists prevent their work from being used by AI art generators. Some of the most popular versions of the new technologies have been Glaze, Kudurru, and Nightshade.
Most AI image generators, like Midjourney, Stable Diffusion, and Dall-E, work by searching databases containing text-image pairs, recognizing patterns between these pairs, and then piecing together a new image based on the patterns they find. Much like how an AI chat bot, such as ChatGPT, puts words together by looking at what’s been written online, looking for patterns in the words it finds, and stringing these words together based on what’s most likely to come next, AI art generators search for patterns and create the image most likely to be associated with a specific prompt.
In an effort to combat these techniques, some services, like Glaze, make minor pixel-by-pixel changes to an image in order to make it more difficult for an AI art service to be able to recognize the patterns in a work. Other services, like Kudurru, stop the problem on the generation side of things by identifying the IP addresses of computers sifting through the internet in an attempt to generate an image, and then sending them something entirely unwanted instead.
Nightshade, which was developed by Ben Zhao, a computer science professor at the University of Chicago, works differently from each of the aforementioned technologies. Nightshade targets the way AI art generators find patterns in text-image pairs by making small tweaks in images to create false text-image pairs. Then, when an AI attempts to create a new image based on these false pairs, it creates an unsatisfactory image.
While these technologies offer a fairly effective way for artists to protect their work from being used by AI at the moment, there are still some concerns about the effectiveness of these technologies over time. Instead of seeing the AI art industry fizzle out due to these new challenges, we may see it simply implement new strategies to get around them.
That being said many artists are currently celebrating these new tools and the protections they afford. As AI becomes increasingly popular as an art generation tool, many artists feel that the nonconsensual use of their work as source material for AI image generation constitutes copyright infringement. So, as lawsuits regarding this issue continue to be fought in courts, any step artists could take to deter these AI services is considered a win by the community.
Leave a Reply