Live data from Hacker News

Nightshade: An offensive tool for artists against AI art generators

nightshade.cs.uchicago.edu

611–620 of 710 posts

Re: Nightshade: An offensive tool for artists against AI art generators

#611
post #68

Earlier quoted context omitted.

I did not say they were exclusively that; I said they were that. Once we as a society have implemented a good way for the artists whose work powers these machines to survive, you can feel good about using them. Until then, frankly, you're doing something immoral by paying to use them.

By this logic we ought to start lynching artists, why they didn't care about all of those who lost their jobs making pigments, canvasses, pencils, brushes etc etc

Why on earth would you choose the word 'lynch' here? At worst I'm suggesting mild disapproval and regulation, not historically racist mob violence.

Re: Nightshade: An offensive tool for artists against AI art generators

#612

Earlier quoted context omitted.

The screenshots you sent in [1] are inference, not training. You need to get a Nightshaded image into the training set of an image generator in order for this to have any effect. When you give an image to GPT-4V, Stable Diffusion img2img, or anything else, you're not training the AI - the model is completely frozen and does not change at all[0]. I don't know if anyone else is still scraping new images into the genera…

Hey you know what might not be AI generated post-2021? Almost everything run through Nightshade. So given it's defeated, which is pretty likely, artists have effectively tagged their own work for inclusion.

Modern generative image models are trained on curated data, not raw internet data. Sometimes the captions are regenerated to fit the image better. Only high quality images with high quality descriptions.

Re: Nightshade: An offensive tool for artists against AI art generators

#613

Earlier quoted context omitted.

>the artists should still have the right to refuse their art being studies. No, that right doesn't exist. If you put your work of art out there for people to see, people will see it and learn from it, and be inspired by it. It's unavoidable. How could it possibly work otherwise? Artist A: You studied my work to produce yours, even when I asked people not to do that! Artist B: Prove it. What kind of evidence or argume…

The right to not use my things exists everywhere, universally. Good people usually ask before they use something of someone else's, and the person being asked can say "no." How hard is that to understand? You might believe they don't have the right to say "no," but they can say whatever they want. Example: If you studied my (we will assume "unique") work and used it without my permission, then let us say I sue you. A…

> The right to not use my things exists everywhere, universally.

For physical rival [1] goods, yes. Not necessarily the same for intangible non-rival things (e.g. the text of a book, not the physical ink and paper). Copyright law creates a legal right of exclusive control over creative works, but to me there isn't a non-economic-related social right to exclusive control over creative works. In the US, fair use is a major limit on the legal aspect of copyright. The First Amendment's freedom of expression is the raison d'être of fair use. Most countries don't have a flexible exception similar to fair use.

> OR, I might agree it is fair use and not sue you. However, that weakens my standing on my copyright, so it's better for me to sue you

No, choosing not to sue over a copyrighted work doesn't weaken your copyright. It only weakens the specific case of changing your mind after the statute of limitations expires. The statute of limitations means that you have a time limit of some number of years (three years in the US) to sue, with the timer starting only after you become aware of an instance of alleged infringement. Copyright is not like trademark. You don't lose your copyright by failing to enforce it.

Furthermore, even though the fair use right can only be exercised as an affirmative defense in court, fair use is by definition not copyright infringement [3]:

> Importantly, the court viewed fair use not as a valid excuse for otherwise infringing conduct, but rather as consumer behavior that is not infringement in the first place. "Because 17 U.S.C. § 107[9] created a type of non-infringing use, fair use is 'authorized by the law' and a copyright holder must consider the existence of fair use before sending a takedown notification under § 512(c)."[1]

(Ignore the bracket citations that were copied over.)

[1] https://en.wikipedia.org/wiki/Rivalry_(economics)

[2] https://www.law.cornell.edu/uscode/text/17/507

[3] https://en.wikipedia.org/wiki/Lenz_v._Universal_Music_Corp.

Re: Nightshade: An offensive tool for artists against AI art generators

#614

Earlier quoted context omitted.

The screenshots you sent in [1] are inference, not training. You need to get a Nightshaded image into the training set of an image generator in order for this to have any effect. When you give an image to GPT-4V, Stable Diffusion img2img, or anything else, you're not training the AI - the model is completely frozen and does not change at all[0]. I don't know if anyone else is still scraping new images into the genera…

Hey you know what might not be AI generated post-2021? Almost everything run through Nightshade. So given it's defeated, which is pretty likely, artists have effectively tagged their own work for inclusion.

Why wouldn't an artist just generate AI spam and Nightshade it?

Re: Nightshade: An offensive tool for artists against AI art generators

#616
post #584
post #566

Earlier quoted context omitted.

Well then I could have been much clearer because I meant something like the latter. An ML model can neither have nor be in breach of copyright so any discussion about how it works, and how that relates to how people work or “learn” is besides the point. What actually matters is firstly details about collation of source material, and later the particular legal details surrounding attribution. The last part involves br…

> and how that relates to how people work or “learn” is besides the point It is important (for the training and generation stages) to distinguish between whether the model copies the original works or merely infers information from them - as copyright does not protect against the latter. > The first part, collation of source material for training is emphatically not unexplored legal or moral territory. Similar to as…

Listen, most website and book-authors want to be indexed by google. It brings potential audience their way, so most don’t make use of their _right_ to be de-listed. For these models, there is no plausible benefit to the original creators, and so one has to argue they have _no_ such right to be “de-listed” in order to get any training data currently under copyright.

Re: Nightshade: An offensive tool for artists against AI art generators

#617
post #169

Earlier quoted context omitted.

That's not really how consent works. I hope this is a special exception you've made, rather than your general approach towards interacting with your fellows.

That is how consent works.

If my body is in a public space you have the right to force me to have sex with you, I guess? That's your logic after all.

Re: Nightshade: An offensive tool for artists against AI art generators

#618
post #48

This seems to introduce levels of artifacts that many artists would find unacceptable: https://twitter.com/sini4ka111/status/1748378223291912567 The rumblings I'm hearing are that this a) barely works with last-gen training processes b) does not work at all with more modern training processes (GPT-4V, LLaVA, even BLIP2 labelling [1]) and c) would not be especially challenging to mitigate against even should it become…

Yeah. At worst a simple img2img diffusion step would mitigate this, but just eyeballing the examples, traditional denoisers would probably do the job? Denoising is probably a good preprocessing step anyway.

It’s a common preprocessing step and I believe that’s how glaze (this lab’s previous work) was defeated.

Re: Nightshade: An offensive tool for artists against AI art generators

#619

Earlier quoted context omitted.

I guess the idea is that the model trainers are ignorant of this and wouldn't know to preprocess/wouldn't bother? That's actually quite plausible.

> I guess the idea is that the model trainers are ignorant of this Maybe they're ignorant of it right up until you announce it, but then they're no longer ignorant of it.

Right, but they aren't necessarily paying attention to this.

I am not trying to belittle foundational model trainers, but a lot goes on in ML land. Even groups can't track every development.

Re: Nightshade: An offensive tool for artists against AI art generators

#620

I think the artists need to agree to stop making art altogether. That ought to get people’s attention. Then the AI people might (be socially pressured or legally forced to) put their tools away.

No, they'll just demand that artists produce more art so they can continue scraping, because if you work in tech you're allowed to be entitled, you're The Face Of The Future and all you're trying to do is Save The World, all these decels are just obstacles to be destroyed.
Post reply on HN