Live data from Hacker News

Nightshade: An offensive tool for artists against AI art generators

nightshade.cs.uchicago.edu

511–520 of 710 posts

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

#511

[flagged]

This is a tired argument; whether or not the diffusion models are "learning", they are a tool of capital to fuck over human artists, and should be resisted for that reason alone.

Exactly. Artists should drop the pretentious philosophical bumbling and accept what this is, a fight for their livelihood. Which is, in every sense, completely warranted and good.

Putting blame on the technology and trying to limit public access to software will not go anywhere. Your fight for regulation needs to be with publishers and producers, not with the teen trying to make a cool new wallpaper or the office-man trying to make an aesthetic powerpoint presentation.

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

#512
I love it. This undermines the notion of ground truth. What separates correct information from incorrect information? Maybe nothing! I love how they acknowledge the never ending attack versus defense game. In stark contrast to "our AI will solve all your problems".

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

#513
post #363

Earlier quoted context omitted.

Then let's use a more precise term that is also present in law: monopoly. You can't monopolize an idea. Copyright law is a prescription, not a description. Copyright law demands that everyone play along with the lie that is intellectual monopoly. The effectiveness of that demand depends on how well it can be enforced. Playing pretend during the age of the printing press may have been easy enough to coordinate, but it…

Trying to make sense of the above comment is difficult. > Copyright law demands that everyone play along with the lie that is intellectual monopoly. Saying "lie" suggests willful deception. Perhaps you mean "socially constructed"? Combined with "playing pretend" makes it read a bit like a rant. > Then let's use a more precise term that is also present in law: monopoly. Ok, in law and economics, the core idea of monop…

The best attempt at thought monopoly I can think of is religion. Even that is a general failure: no single religious narrative has ever stood constant. They have all evolved through the participation of their adherents. There is no one true Christianity: there are hundreds.

I most certainly do mean to call out copyright as willful, but it's not a deception, at least not a successful one: everyone knows it is false. That's why it's enforced by law! Instead of people being deceived, people must instead pretend to be so. Each of us must behave as if the very concept of Micky Mouse is immortal and immutable; and if we don't, the law will punish accordingly.

Every film on Netflix, every song on Spotify, etc. can obviously be copied any number of times by any number of people at any place on Earth. We are all acutely aware of this fact, but copyright tells us, "Pretend you can't, or get prosecuted."

So is it truly effective? Millions of people are not playing along. Millions of artists are honestly trying to participate in this market, and the market is failing them. Is that because we need more people to play along? Rightsholders like the MPAA say that piracy is theft, and that every copy that isn't paid for is a direct cost to their business. How many of us are truly willing to pretend that far?

What if we all just stopped? Would art suddenly be unprofitable for everyone, including the lucky few who turn a profit today? I don't believe that for a second.

The only argument I have ever heard in favor of copyright is this: Every artist deserves a living. I have seen time and time again real living artists fail to earn a living from their copyright. I have seen time and time again real living artists share their work for free, choosing to make their living by more stable means.

Every person living deserves a living. Fix that, and we will fix the problem copyright pretends to solve, and more.

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

#514
post #34

Earlier quoted context omitted.

great comment! imagine being a photographer that takes decades to perfect their craft. sure another student can study and mimic your style. but it's still different than some computer model "ingesting" vast amount of photos and vomiting something similar for $5.99 in aws cpu cost so that some prompt jockey can call themselves an AI artist and make money off of other peoples talent. i get that this is cynical and does…

Like 99.9% of the art the common people care about is Darth Vader and Taylor Swift and other pop culture stuff like that. These people literally don’t care what your definition of what is and isn’t art is, or how it’s made, they just want a lock screen wallpaper of themselves fighting against Thanos on top of a volcano. The argument of “what is art” has been an academic conversation largely ignored by the people actu…

This response should be gilded

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

#515
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…

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…

Correct me if I'm wrong but I understand image generators as relying on auto-labeled images to understand what means what, and the point of this attack to make the auto-labelers mislabel the image, but as the top-level comment said it's seemingly not tricking newer auto-labelers.

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

#516
post #447

Earlier quoted context omitted.

"Sorry, this is not art, is AI generated trash."

Who cares? With AI, you don't need art school. AI is making humanities redundant, and people are too proud to admit it. I can't believe how many people are not in awe of the possibilities of AI art, so it's great to see AI disturbing the cynics until they learn. Not everything is political, but I'll let them have this one.

> With AI, you don't need art school. AI is making humanities redundant, and people are too proud to admit it.

If you think this is true, you've never understood art. Art is a human endeavour fundamentally. What ML produces is not art.

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

#517
post #175

Huge market for snake oil here. There is no way that such tools will ever win, given the requirements the art remain viewable to human perception, so even if you made something that worked (which this sounds like it doesn’t) from first principles it will be worked around immediately. The only real way for artists or anyone really to try to hold back models from training on human outputs is through the law, ie, levera…

My guess. Is that at some poi t of time You will not be able to use any generated image or video in commercial. Because of 100% copyright claim for using parts of copyrighted image. Like YouTube those days. When some random beeps matches with someone music...

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

#518
post #175

Huge market for snake oil here. There is no way that such tools will ever win, given the requirements the art remain viewable to human perception, so even if you made something that worked (which this sounds like it doesn’t) from first principles it will be worked around immediately. The only real way for artists or anyone really to try to hold back models from training on human outputs is through the law, ie, levera…

> Huge market for snake oil here. This tool is free, and as far as I can tell it runs locally. If you're not selling anything, and there's no profit motive, then I don't think you can reasonably call it "snake oil". At worst, it's a waste of time. But nobody's being deceived into purchasing it.

If this is a danger from "snake oil" of this type, it'd be from the other side, where artists are intentionally tricked into believing that tools like this mean that AI isn't or won't be a threat to their copyrights in order to get them to stop opposing it so strongly, when in fact the tool does nothing to prevent their copyrights from being violated.

I don't think that's the intention of Nightshade, but I wouldn't put past someone to try it.

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

#519

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.

It is a great shame that we have come to a no-win situation for artists when VCs are virtually unable to lose.

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

#520
post #267
post #175

Huge market for snake oil here. There is no way that such tools will ever win, given the requirements the art remain viewable to human perception, so even if you made something that worked (which this sounds like it doesn’t) from first principles it will be worked around immediately. The only real way for artists or anyone really to try to hold back models from training on human outputs is through the law, ie, levera…

The level of claims accompanied by enthusiastic reception from a technically illiterate audience make it sound, smell, and sound like snake oil without much deep investigation. There is another alternative to the law. Provide your art for private viewing only, and ensure your in person audience does not bring recording devices with them. That may sound absurd, but it's a common practice during activities like having…

The thing is people want the benefits of having their stuff public but not bear the costs. Scraping has been mostly a solved problem especially when it comes to broad crawling. Put it under a login, there, no more AI "stealing" your work.
Post reply on HN