[limitedsupply]

Another counterargument is that these models do not generate anything without a prompt. So in a way it’s like a very smart photoshop: if you, a designer, use tools to create a character that resembles Mickey Mouse - this is on you, not the tool. It’s going to be interesting to hear this being argued in court.

[williamcotton]

Exactly. The model only produces copyright protected material when the operator of the model directs the sampler towards the part of the latent space that contains that material.

The user is liable. The user is also liable for using a photocopier to create and sell copies of someone else’s work. Anyone can photocopy anything they want and plaster their own walls with that work. In all cases the manufacturer of the photocopier is not liable…

Now, if these tools consistently plagued users with being liable for copyright violations then these tools become more trouble than they are worth.

[karpierz]

A password-protected zip file containing Mickey Mouse photos also doesn't produce copyrighted material without the right prompt. Yet, the distributor of the zip file is definitely responsible for copyright violation.

And to preempt a likely counterargument, intent is not a required factor for establishing copyright violation.

[williamcotton]

> intent is not a required factor for establishing copyright violation.

What you mean to say is that if someone just so happens to write a story that is significantly similar to another work that their intent doesn't matter... most importantly their intent is unknowable.

This doesn't mean that courts will not be interested in the intended use of specific mechanical tools such as VCRs, computers, and statistical models.

Let's tease apart this "zip file containing Mickey Mouse photos" because it seems to be from where your errors in reasoning stem.

Here is a website that gives you instructions on how to draw Mickey Mouse:

https://courses.cs.duke.edu/cps004/fall01/examples/mickey/St...

These instructions are not violation of copyright because they draw a picture of Mickey Mouse. Copyright does not protect the mere existence of an image. Copyright gives a monopoly on distributing copies of that image to the rights holder. Not the idea of the image, not the instructions that could lead to the image, not the 1s and 0s that make up the image.

As I've mentioned before, a t-shirt with the 1s and 0s that make up the zip file that contains those photos is not in violation of copyrights on those images. You'll notice that CafePress still has t-shirts with the DeCSS code for sale: https://www.cafepress.com/+,954530397

Imagine a mathematical formula that when plotted in polar coordinates draws Mickey Mouse. The mathematical formula is a fact and it cannot be copyrighted. There's another step required for copyright infringement: Someone renders the image and distributes copies of the rendered image.

[williamcotton]

I mentioned VCRs specifically because in 1976 Universal Studios sued Sony over the claims that because Sony manufactured a device that could be used for copyright infringement the company was liable for infringement committed by consumers of Betamax.

https://en.wikipedia.org/wiki/Sony_Corp._of_America_v._Unive....

Take a note of the majority opinion in the Supreme Court ruling:

> The question is thus whether the Betamax is capable of commercially significant noninfringing uses ... one potential use of the Betamax plainly satisfies this standard, however it is understood: private, noncommercial time-shifting in the home.[7] [...] [W]hen one considers the nature of a televised copyrighted audiovisual work... and that time-shifting merely enables a viewer to see such a work which he had been invited to witness in its entirety free of charge, the fact... that the entire work is reproduced... does not have its ordinary effect of militating against a finding of fair use.

Stable Diffusion is also capable of commercially significant non-infringing uses. It is to be expected that the lower courts will continue to follow the precedence set by the Supreme Court that there is a clear distinction between a mechanical device capable of making copies and the distribution of those copies.

To put in the context of Stable Diffusion, this is why legal experts keep saying there are two things to consider: The model, which is almost guaranteed to be protected by fair use, and the outputs of the model, in which the users of the model are liable for any copyright infringement.

[karpierz]

> Copyright gives a monopoly on distributing copies of that image to the rights holder.

Or derivative works.

> not the instructions that could lead to the image,

And yet torrent sites have been found to violate copyright, despite not hosting any of the copyrighted content.

> not the 1s and 0s that make up the image.

Now this is plainly false. I cannot start selling a Mickey Mouse JPEG stripped straight from Disney's site.

> You'll notice that CafePress still has t-shirts with the DeCSS code for sale: https://www.cafepress.com/+,954530397

1) This is covered by a trade secrets law, and not copyright.

2) It still being for sale doesn't mean it doesn't violate copyright; only that the copyright holder hasn't litigated to stop it. Put differently, getting away with something doesn't make it legal.

Look, I think Stable Diffusion is pretty neat. But all of this "it's just math"/"it's just data"/"it also creates non-copyrighted images" stuff doesn't cut it as far as the law is concerned. If your point is that being in possession of the model isn't violating copyright, sure.

I'm talking about distributing or otherwise providing a service that uses the model to generate images. If they figure out how to get it to stop generating training data, they'll be fine. But with existing approaches, that'll be nigh impossible.

[williamcotton]

> > not the 1s and 0s that make up the image.

> Now this is plainly false. I cannot start selling a Mickey Mouse JPEG stripped straight from Disney's site.

You can absolutely go to the Disney website right now, download any JPEG, convert that image to a series of 1s and 0s, and then print that series of 1s and 0s on a t-shirt that you sell on your website.

Once those 1s and 0s are in a new non-digital medium they are no longer a copy of the image. Historically, the digital copy is a special case because practically there is no difference between the "expression" of a byte sequence in a digital medium and the "idea" of a byte sequence in a digital medium.

A t-shirt is not a digital medium so the rights holders don’t get to extend their monopolization of speech to a screen print of an unintelligible pattern of 1s and 0s.

[karpierz]

> A t-shirt is not a digital medium so the rights holders don’t get to extend their monopolization of speech to a screen print of an unintelligible pattern of 1s and 0s.

Can you point me to a ruling where this form of expression is successfully defended via fair use?

Seems unlikely that a judge would agree that a simple transformation between "digital" vs "non-digital" medium avoids copyright infringement (ex: if I produce a digital image, I definitely maintain copyright over digital and non-digital forms of that image; you can't just start printing and selling posters of my image).

You could try to argue that the change of medium is transformative, but that's not a generic argument for "digital" vs. "non-digital"; that's particular to the circumstances of the case.

[williamcotton]

Briefly, more to come:

What are you missing about the fact that a t-shirt with Mickey Mouse on it and a t-shirt with a pattern of 1s and 0s are completely different. Is anyone buying and wearing the 1s and 0s shirt instead of buying and wearing a Mickey shirt?

You can’t say the same thing about digital copies! Is anyone buying the 1s and 0s of a JPEG in a digital medium and expecting to just have a bunch of 1s and 0s?

BTW, if anyone wants access to those 1s and 0s they only need to go to the Disney website to download the JPEG… they don’t need to buy my 1s and 0s shirt.

You’re just not thinking about this in legal terms: who has wronged who and how did they wrong them. These lawsuits are not criminal, not public wrongs, but rather private wrongs. People must show that there are concrete damages against their specific property!

[williamcotton]

I’m having a hard time figuring out what it is you’re missing here but I’m enjoying that I get to hone my rhetorical chops with this argument.

Finding the relevant case law is going to take me a little bit of time and it will probably require a bit of exposition so please be patient!

[williamcotton]

> If they figure out how to get it to stop generating training data, they'll be fine.

How does that change anything at all? Take any of the plaintiffs in the Stable Diffusion lawsuit. Let’s say that their work was left out of the training data or if it was allowed to remain that the tool never reproduced training data… is Stable Diffusion any more or any less useful? Is literally anyone using Stable Diffusion with the intent to reproduce one of the plaintiff's copyright protected works?

[karpierz]

> Let’s say that their work was left out of the training data or if it was allowed to remain that the tool never reproduced training data… is Stable Diffusion any more or any less useful?

This doesn't really have much to do with copyright.

> Is literally anyone using Stable Diffusion with the intent to reproduce one of the plaintiff's copyright protected works?

Intent isn't really a factor in terms of damages for copyright infringement. But I assume we'll figure out the extend of the damages if/when damages are awarded, since that's when they measure it out. Granted, I expect the liable parties to settle far before it gets to that; the payout seems cheap (~100k - 1m) relative to the risk (penalties for copyright infringement are per unit, and it's easy to hit millions of units on the internet).

[williamcotton]

And yes, a lawyer for the defense will show that, unlike a zip file of photos that can ONLY be copies of protected works and that anyone distributing such a zip file cannot reasonably say that they are expecting non-infringing works, with Stable Diffusion there is basically no one expecting to get copies of any of the plaintiffs protected works.

Like, literally no one who made or uses SD knows or cares about the plaintiffs, is interested in their drawing style, and even remotely interested in any of their individual protected works. If I were the defendants I would train a model without any of the plaintiffs’ work and then show the court that the results for an astronaut riding a horse are no worse without their works.

And no, some lawyer is not going to scream out “objection, intent has nothing to do with copyright”, because, again, all that is meant by the intent is this: if Stable Diffusion output an image but your intent was not to copy another image… well too bad, you’re still infringing… this all happens where the liability for disseminating the copy and foregoing any fair use is on the person using the tool. Is using the tool worth the risk? Maybe not!

The intentions of the toolmakers are very important to the courts. A tool like a torrent website that does literally nothing to police their content and even actively promotes piracy is going to be treated differently than publicdomaintorrents.info.

[williamcotton]

> > Let’s say that their work was left out of the training data or if it was allowed to remain that the tool never reproduced training data… is Stable Diffusion any more or any less useful?

> This doesn't really have much to do with copyright.

It is not specific to copyright but any private wrong heard by the courts needs a clearly defined plaintiff, defendant and property in question.

How can the plaintiffs accuse anyone of infringing on their copyrights when no one has used the tool to produce the protected works? Which of their specific works was infringed upon? Is that specific work a requirement for the tool to function?

The plaintiffs seem to be “everyone” and their property “every single digital image”, which is legally incoherent. How exactly would someone reasonably ask permission for every single digital image? Is everyone entitled to part of the settlement or just the plaintiff’s and their lawyers?

[williamcotton]

Intent is absolutely a factor when accessing the tool. Intent is not a factor when accessing the infringing work.

The intentions of Sony when they manufactured the Betamax VCR were absolutely taken into consideration in the judgment.

[williamcotton]

> But all of this "it's just math"/"it's just data"/"it also creates non-copyrighted images" stuff doesn't cut it as far as the law is concerned.

What you mean is that you would prefer if Stable Diffusion was not considered fair-use by our legal system.

However, based on my more than cursory understanding of copyright law in the US, it seems more likely that the courts will find these tools fair-use than otherwise.

[karpierz]

> What you mean is that you would prefer if Stable Diffusion was not considered fair-use by our legal system.

I think you misunderstand me. Despite my feelings about Stable Diffusion and laws around copyright, I believe the law will find in favour of the artists, assuming they have the funds/will to see it through. The law unfortunately doesn't really care how I feel about it.

Granted, they're up against some fairly large corporations that have both time and money to stall for a while, so I don't know how far their lawsuit will go.

[williamcotton]

> And yet torrent sites have been found to violate copyright, despite not hosting any of the copyrighted content.

BitTorrent itself is legal. The torrent sites that were shut down were found to have basically no non-infringing utility. There were and still are other torrent sites that have non-infringing utility which are of course allowed to exist.

[karpierz]

> BitTorrent itself is legal.

Yes, and so is Winzip, and the code that trains Stable Diffusion. But that doesn't have much to do with the resulting data that is used and generated by those files.

> There were and still are other torrent sites that have non-infringing utility which are of course allowed to exist.

Can you point me to a torrent site that withstood a legal challenge for copyright infringement?

[williamcotton]

https://www.publicdomaintorrents.info/

[limitedsupply]

The important distinction is that the model does not contain a copy of a copyrighted material. It contains data trained on copyrighted material (like our brains).

[karpierz]

You're making a distinction that isn't reflected in copyright law.

If I take a JPEG of Mickey Mouse and then turn it into a PNG, it's not a "copy", as the bits are different. But it still contains copyrighted material.

You can try to argue that the bits of PNG itself aren't an image of Mickey Mouse, but rather the algorithm that reads the PNG produces an image of Mickey Mouse. But that isn't really a relevant distinction in so far as copyright is concerned.

In addition, this statement is false:

> The important distinction is that the model does not contain a copy of a copyrighted material.

It has been shown repeatedly that the model produces copies of training data. The copies are of course not stored as JPEGs/PNGs in the model, but they are retrievable from the model, given the correct password (prompt).

[limitedsupply]

Could you provide evidence of your last statement? I haven’t seen these models produce actual copies of any art (can’t imagine that’s an option in general).

These models do not contain copies. One way to describe the data is they contain a statistical breakdown of the artwork, which is substantially different from a JPEG -> PNG conversion you mention.

[karpierz]

> Could you provide evidence of your last statement? I haven’t seen these models produce actual copies of any art (can’t imagine that’s an option in general).

Here's a whole paper, complete with citations:

https://arxiv.org/pdf/2212.03860.pdf

> These models do not contain copies. One way to describe the data is they contain a statistical breakdown of the artwork, which is substantially different from a JPEG -> PNG conversion you mention.

I don't understand the distinction you're making. What legally separates a "statistical breakdown" representation from a zip file representation, JPEG representation, PNG representation?

[limitedsupply]

As I anticipated, you are referencing research that does not show exact copies being generated by Stable Diffusion. Do "semantically equivalent" images infringe on copyright? I would argue that they do not. We will see how this plays out in court.

Food for thought: if I write instructions for generating an SVG of a black square, does my program contain copyrighted material (Malevich's Black Square)? You and I could argue about that, but you will probably quote more research that disproves your own point. So let's skip that.