Glaze and Nightshade "AI Poison Pill"

suffo85suffo85 Posts: 177

I ran across these two pieces of software specifically developed to disrupt generative AI, and those training their models on copyrighted art without permission.  They're both free though I suspect over time we will probably see paid alternatives spring up.

Since I don't run any of my own LLMs, I'm unable to test them, but in theory it should work.  Both pieces of software come from an academia source, The University of Chicago.

Merely passing along this information for those interested in protecting their art styles, and those interested in being a little more than simply passive about it can opt for Nightshade to effectively poison a large learning model over time.

 

Glaze is described by the University of Chicago as "a system designed to protect human artists by disrupting style mimicry. At a high level, Glaze works by understanding the AI models that are training on human art, and using machine learning algorithms, computing a set of minimal changes to artworks, such that it appears unchanged to human eyes, but appears to AI models like a dramatically different art style. For example, human eyes might find a glazed charcoal portrait with a realism style to be unchanged, but an AI model might see the glazed version as a modern abstract style, a la Jackson Pollock. So when someone then prompts the model to generate art mimicking the charcoal artist, they will get something quite different from what they expected." [source]

Nightshade is described as "an offense tool to distort feature representations inside generative AI image models. Like Glaze, Nightshade is computed as a multi-objective optimization that minimizes visible changes to the original image. While human eyes see a shaded image that is largely unchanged from the original, the AI model sees a dramatically different composition in the image. For example, human eyes might see a shaded image of a cow in a green field largely unchanged, but an AI model might see a large leather purse lying in the grass. Trained on a sufficient number of shaded images that include a cow, a model will become increasingly convinced cows have nice brown leathery handles and smooth side pockets with a zipper, and perhaps a lovely brand logo." [source]

 

Visit University of Chicago's Glaze Website - https://glaze.cs.uchicago.edu/

Visit University of Chicago's Nightshade Website - https://nightshade.cs.uchicago.edu

 

Personally, I'll be using both moving forward.

Feel free to discuss your thoughts on the matter. smiley

Comments

  • Phoenix1966Phoenix1966 Posts: 1,672

    Thanks for sharing the links.

  • xyer0xyer0 Posts: 5,937

    Sounds like a cool idea if it's legit, I mean, the University of Chicago.

  • doubledeviantdoubledeviant Posts: 1,142
    Spicy. Thanks!
  • TorquinoxTorquinox Posts: 3,336
    edited May 11

    Can repeat what I've already said:

    Understand that Glaze and Nightshade are basically ai products being used to fight ai generators. The download for glaze and nightshade is over 10Gb - not trivial, sort of small for a contemporary video game. You must also install the NVidia CUDA Toolkit, and have an acceptable NVidia graphics card for it to work. Even then, what it does is somewhat limited - you can't really apply both Glaze and Nightshade to an image, at least not yet. And the software requires time and computing resources to re-render your image with the defenses. A lot of people would have to be doing all that to make it truly effective, and it's not clear to me that it is truly effective.

    Post edited by Torquinox on
  • TorquinoxTorquinox Posts: 3,336

    Also, it is recommended to re-treat images with each new release. I downloaded and installed on Glaze 1.1. Now we're at Glaze 2.0 - The researchers say it's better, faster, etc. Seems like it's another treadmill and you'll have to reprocess your images.

    Of course, you don't actually control your images once they're on the web - Could be copies of them in other places that you don't control, maybe don't even know about.

Sign In or Register to comment.