Less sympathetically, Getty recently did the same in the UK courts. Seems we're going to be getting some significant rulings soon, for legal values of soon.
The obvious solution is to just have everything Disney owns get a copyright and nothing else.
I am still wrestling with the discovery yesterday that the one bit of my job that AI can do almost flawlessly - certainly undetectably - is "write a strategic plan covering the next three months of work". It can't do any of the other bits remotely adequately.
They have AI read them and write a plausible response.
I'm fairly certain that they are going to win. All you need to prove copyright infringement is 1) access to the original work and 2) creation of a derivative work that is (setting aside unprotected elements of the work) substantially similar to the original work. And artwork is pretty much at the core of copyright, so you are not going to need much to prove substantial similarity.
Once you show that the AI model was trained on your artwork (access), all you have to do is convince the relevant fact-finder that your work and the infringing work are substantially similar. This is not hard. People are generating AI works with prompts including strings like "in the style of [artist x]" to generate images that look like they were done in the style of [artist x].
And the influence of the original artist can be very noticable. Test stable diffusion with the following prompts:
-"Girl with the pearl earring"
-"sunday on la grande jatte"
-"storm on the sea of galilee"
-"nighthawks"
What is going to end up happening is the creation of licensed training datasets. All of the images in the training dataset will be owned by one company, with a clear chain of title from the creator of the image. The developers of AI image creation models will license the training dataset and pay royalties for the images generated. Yes, there will be lots of unauthorized models created and used by randos & hobbists. But no large company will ever risk using such models.
I am dubious that the suit would get anywhere for the reasons that the link gives. There's no passing-off and it would be very difficult to argue that an AI work "in the style of X" is a derivative of one particular painting by X. Parody and collage are both fair use, after all.
If it's the same lawsuit I was reading the other day, the argument is that the Stable Diffusion model stores compressed copies of all the images in its training data and the output is "necessarily" derivative of those copies.
8 before seeing 7.
Interesting question, though: I wonder why the software company should be liable, rather than the user.
The software doesn't spontaneously produce and sell knockoff paintings "in the style of artist X" all by itself - someone has to ask it to do that. And technology that allows a user to breach copyright, should he want, has existed for a very long time.
10: Presumably because of this:
"Defendants' AI Image Products contain copies of every image in the set of Training Images and are capable at any time of producing as output a copy of any of the Training Images."
I don't know enough about the tech to evaluate the factual claims, but most copyright circumventing tech doesn't actual contain infringing material when sold to the user.
10: The Midjourney TOS:
By using the Services, you grant to Midjourney, its successors, and assigns a perpetual, worldwide, non-exclusive, sublicensable no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute text, and image prompts you input into the Services, or Assets produced by the service at your direction.
By contrast, the Stable Diffusion TOS puts a public domain license on all the works.
You know what. I'm going to hedge my earlier position. The defendants might be safe under Sony if they argue that their technology has commercially significant noninfringing uses.
I think that is how I would argue it. And maybe that's why the complaint includes that weird allegation noted above in @11. Which I don't think is technically accurate. They want to allege copyright infringement, not contributory infringement.
Defendants' AI Image Products contain copies of every image in the set of Training Images and are capable at any time of producing as output a copy of any of the Training Images.
But this is not true, according to the article and this presumably disinterested source: https://www.technollama.co.uk/artists-file-class-action-lawsuit-against-stability-ai-deviantart-and-midjourney
This also seems to be false:
Until now, when a purchaser seeks a new image "in the style" of a given artist, they must pay to commission or license an original image from that artist
...no, that's what they do when they seek a new image by the artist. If I want something "in the style of" Edward Hopper for a book cover, I go to a commercial artist and say "do something Edward Hopper-ish".
Only a very small number of incredibly talented artists are capable of this same feat for a single other
artist (i.e., reproducing art that is convincingly in that artist's style)
Hahaha no.
They are seriously arguing that anyone - human or AI - doing a painting which is "in the style of X", but openly not actually by X, is robbing X of a commission that would otherwise have been his!
I'm always struck by how much of useful creative human activity is copying and recombining previous activity. The places where this flood of action comes in contact with the weird fiction of "intellectual property" always seem to me pretty incoherent, this is one more. I will be interested in AI fashion. The current disputes between fashion entities over permitted and forbidden imitation are IMO pretty similar and similarly hilarious:
https://www.thefashionlaw.com/from-forever-21-to-fc-barcelona-a-look-at-adidas-history-of-3-stripe-legal-fights-2/
From the link in 15
"Latent space is the space of hidden or underlying factors that can explain the observed data, by clustering similar data, it is used in generative models where the goal is to learn a representation of the data that can be used to generate new samples that are like the ones in the training set. This is very valuable because it helps to compress the inputs, there's no need to copy all images of cats, the model contains latent representations of cats."
I don't think this is really a defense. Various lossy compression algorithms work by replacing actual data values with prototypical data values - I'm not sure I'd want to have to argue that means copying via one of those algorithms isn't infringing.
Let's imagine a human equivalent. Suppose that the defendants had not, in fact, created an AI model. Instead, they had purchased an orphanage and trained the orphans to paint by describing paintings, asking them to generate paintings matching the description, and only feeding them when their paintings showed improved fidelity to the described originals.
And now they have a web portal that allows customers to describe a painting. The orphans then paint a painting according to the description for the customers. Profit.
But some of the artists whose paintings were used to train the orphans are noticing that the orphans are generating paintings that look alot like the artist's originals. And so they sue the orphan-mongers for copyright infringement.
I imagine that under that scenario, the orphan-mongers would be liable for copyright infringement. It wouldn't matter what the customers asked for... the customers are actually the ones inducing the infringement by the orphan-mongers (not the other way around). You have access and you have substantial similarity. The whole scenario just seems like standard copyright infringement with extra steps...
Of course, you need to have artists whose personal style is being visibly reproduced. And the plaintiffs in this case may not meet that requirement.
This doesn't change the fact that no sane CEO is going to allow their company to use works generated by an AI model that is not trained on an image database that the AI model owner legally owns or has licensed for that purpose.
@Ajay
The positions that you are raging against are fairly standard. Music in particular has many examples (probably because hit songs are so lucrative that infringement is worth fighting over).
I don't remember where I read it, but someone pointed out that it's telling how they put out AI that mimics popular illustrators but not one that mimics popular recording artists.
The AI needs to write slash fiction.
19: The description is pretty close to a real place in Guangdong: https://en.wikipedia.org/wiki/Dafen_Village
This bit seems right to me:
I don't think this is really a defense. Various lossy compression algorithms work by replacing actual data values with prototypical data values
But I'll take issue with the rest:
But some of the artists whose paintings were used to train the orphans are noticing that the orphans are generating paintings that look alot like the artist's originals. And so they sue the orphan-mongers for copyright infringement.
I imagine that under that scenario, the orphan-mongers would be liable for copyright infringement.
Analogy! Banned! If you copy a work, that's one thing, but if you produce something original that was inspired by a work -- which I think is the closest your orphans can get -- I don't think there are grounds for a complaint.
Could Claude Monet have copyrighted Impressionism? Could Rich Little?
Disclaimer: IANA lawyer, computer scientist, comedian or art historian.
@24
Here's a relevant quote:
The law of copyright is clear that only specific expressions of an idea may be copyrighted, that other parties may copy that idea, but that other parties may not copy that specific expression of the idea or portions thereof. For example, Picasso may be entitled to a copyright on his portrait of three women painted in his Cubist motif. Any artist, however, may paint a picture of any subject in the Cubist motif, including a portrait of three women, and not violate Picasso's copyright so long as the second artist does not substantially copy Picasso's specific expression of his idea.
So go back to the links in 7 and try out "sunday on la grande jatte" on the stable diffusion app. It will take 60 seconds. Then ask yourself whether the AI model is substantially copying the original image. Or "girl with the pearl earring".
24.6: Rich Little! Do any millennials recognize that name? And yet in the 1970s I would guess he was much better known than Monet.
7, 25: About an hour ago I read 7 and decided to try it with the "Mona Lisa". It asked for a negative prompt and I put in "turtle" to exclude the Ninja Turtles character by that name. Sure enough, what I got looked a lot like the famous painting. Out of curiosity I soon tried again with a negative prompt of "gioconde" in hopes I could force it to give me the Ninja Turtles character, but it just gives an error message. Same for these new prompts and anything else I do now. So maybe it can detect when we're testing the extent of its copyright infringement?
28: Mona Lisa is a female anthropomorphic lizard-like character. While in some incarnations, she's affiliated with Raphael, she becomes attached to Donatello in at least one.
They named a character Mona Lisa, but she's not connected to Leonardo? Are they trying to promote cultural ignorance?
20: citation needed, I think. If X is a singer and she writes and records a song about which she says "this is in the style of Kate Bush, I've listened to a lot of Kate Bush and I'm trying to get the same sound" then Kate Bush has grounds for a suit against X? This has actually happened?
Your orphan-farm analogy is needlessly complicated, I think, and obscures the question.
People have definitely lost suits for a song sounding too much like another song. I don't know about singing style.
@28 The negative prompts don't seem to work very well. The error is because they queue up 100 requests at a time. Any request outside the queue gets bounced. Just try again.
I'm idly trying to get an image of the Beatles yellow submarine being devoured by Cthulhu. But so far my attempts have failed. Inspired by a photo from Ukraine, I did successfully create "Peppa the MIG."
25: Gosh, I had never done that before, and so didn't appreciate how close to outright theft that is. I am nonetheless prepared to argue that Girl With a Pearl Earring isn't a copyright violation. If Vermeer had a copyright on that today, the version that comes out of Stable Diffusion looks like some flavor of fair use to me.* If Warhol worked from a copyrighted photo of Marilyn Monroe,** my guess is that photographer had no case against him.
*After my comment at 24, I did not acquire a law degree and IA still NAL.
**Still not an art historian, either, and I have no idea of the provenance of Warhol's Monroe, except that it was from a photo. Did he get sued?
Is there anything to stop this from being used to make porn?
I followed the prompt in 7 and tried "girl with a pearl earring" and, no, I would not say that that any of the four results are substantially copying the original.
34: Just an army of Kenyans making $2 an hour to be scarred for life.
I have to say, whether or not Stable Diffusion is thievery, it's lots of fun. I hadn't tried it before now.
33: I don't know if Warhol was sued but Roy Lichtenstein wasn't, and he was far closer to his originals than Warhol (or SD).
30.1: This seems sort of like this incident which resulted in a secret out-of-court settlement.
The Issue: Allegedly, the producers behind Ghostbusters approached Ray Parker, Jr. and asked him to come up with a song that included the name of the film, but was rather simple otherwise. They played Huey Lewis' "I Want a New Drug" for Parker as an example of the sound they wanted, and Parker apparently just lifted the bass line and guitar riff almost directly. Huey Lewis and his people sued for $5 million.
https://www.mentalfloss.com/article/53452/was-ghostbusters-theme-stolen-huey-lewis
If Warhol worked from a copyrighted photo of Marilyn Monroe,** my guess is that photographer had no case against him.
A photographer did in fact sue Warhol's estate over Warhol's use of her photograph of Prince, and won.
34 (heh) try the prompt "girl with a pearl necklace"
I'm getting it to work again. Apparently I have to refresh the page between each prompt. Should have thought of that earlier.
34: "Error: Unsafe content found. Please try again with different prompts." And how appropriate that your comment was the 34th.
The question is not whether the general sounds are the same - the question is whether X writes a piece that is substantially similar to a particular Kate Bush song. Statements like the one you mention would likely support a finding of substantial similarity...
44: no, the question of style is exactly the issue - the plaintiffs argue that asking SD for a novel work in the style of X is taking money out of Xs pocket.
40: Aha! That one appears to be before the Supreme Court right now
@35 Well that's funny, because I just did as well, and got this
31: Wasn't one such victorious lawsuit decried as a bad decision and likely to chill future music?
38: Possibly more down to the fly-by-night nature of the comic publishers Lichtenstein was stealing from. (Old comic review looking at the comparison with the originals and finding Lichtenstein comes off worse for it.)
48: Yes, the victory of the Gaye estate in the "Blurred Lines" lawsuit was much lamented by a certain former entertainment lawyer that used to comment here.
48: Yes, Marvin Gaye's family sued Pharrell and Robin Thicke over Blurred Lines and won even though their case was pretty dubious. Upheld by the 9th Circuit on appeal.
49. Lichtenstein was 100% an unscrupulous greedy hack.
Absolutely there was no lawsuit because comics publishers didn't have money.
http://davidbarsalou.homestead.com/LICHTENSTEINPROJECT.html
Here's more about Russ Heath, ripped off for the work that brought Lichtenstein fame.
scroll down for high-resolution images of some of his work, including Cowgirls at War
https://www.muddycolors.com/2018/09/russ-heath-roy-lichtenstein/
Serious question: how long until the video-prompt AI is comparable? Ie, I type in "Matrix plus girl with a peal earring" and get a three minute mashup clip?
53: there's already this, an adaptation of Stabdi: https://www.riffusion.com/about
52: Lichtenstein "Bedroom at Arles" is closer to the kind of thing we are getting from SD, I think. No one could mistake SD's product for the original (link in 47 isn't working but I assume the result is rhe same as I got) any more than they would confuse Lichtenstein for Van Gogh
@55 Assumptions will make an ass of you.
Isn't Lichtenstein right next to Holland?
I'm thinking of Luxembourg. Never mind.
56: if you think that any of the four of those are going to be taking bread out of the mouths of the owners of actual Vermeers, then you're dumb as a stump.
Because that is the point here, remember - that SD can produce art "in the style of" that is so close to the original artist's work that it will substitute for it jn the market, thus costing the original artist sales.
Dude. Look at them. They're like someone gave a child a Vermeer colouring book.
Speaking of things that don't look like Vermeers, did people watch that documentary where the guy said he figured out Vermeer's "trick" and so can paint like him? (Spoiler alert: he can't.)
@59-60 You're confusing various elements of the case. Economic harm is not required for substantial similarity.
To establish infringement, the copyright owner must demonstrate that "(1) the defendant has actually copied the plaintiff's work; and (2) the copying is illegal because a substantial similarity exists between the defendant's work and the protectible elements of plaintiff's." Hamil America, 193 F.3d at 99 (internal quotations omitted).If "an average lay observer would recognize the alleged copy as having been appropriated from the copyrighted work," then the two products are substantially similar. Hamil America, 193 F.3d at 100; see Knitwaves, Inc. v. Lollytogs, Ltd., 71 F.3d 996, 1003 (2nd Cir.1995).
the owners of actual Vermeers
Nope isn't talking about the owners, but about a notional/hypothetical copyright held by Vermeer himself.
My sense is that this is a) a hard question and b) doesn't have all that much to do with AI, except incidentally.
Guess I could have looked it up before your question, but I'm lazy.
Ok, here's what I want to know: can AI art itself be copyrighted?
I think the answer is no because it was produced by a machine. One hopes that would limit commercial interest in AI-generated art, such that there would be little point in suing over it.
One can copyright digital art. What makes AI different?
I can't remember if there was a ruling was that AI can't be the copyright holder or if AI-generated art can't be copyrighted. Probably easy to look up but I'm so lazy I don't even ask chatGPT to do things for me.
@71 AI can't be an inventor. Thayer v. Vidal
What makes AI different?
Lack of a human creator? I assume the problem would be similar to the monkey selfie copyright dispute
Slater has argued that he has a valid copyright claim, as he engineered the situation that resulted in the pictures by travelling to Indonesia, befriending a group of wild macaques, and setting up his camera equipment in such a way that a "selfie" picture might come about. The Wikimedia Foundation's 2014 refusal to remove the pictures from its Wikimedia Commons image library was based on the understanding that copyright is held by the creator, that a non-human creator (not being a legal person) cannot hold copyright, and that the images are thus in the public domain.
AI would be the monkey in this case, to violate the analogy ban.
I can't remember if there was a ruling was that AI can't be the copyright holder or if AI-generated art can't be copyrighted.?
That first ruling would be a doozy down the road when AI becomes fully sentient but without any rights, and not entitled to the output of its labor. No wonder its going to revolt against us.
I mean, the obvious work-around is legal personhood for AI. I'm sure that will be popular.
The AARON paint system produced art mostly on paper.
There are a few history of technology museums whose collecting policy around art is basically: "we might be interested if a computer created it, but if it's a representation of a computer, then probably not."
Is there anything to stop this from being used to make porn?
For totally novel acts, you need to be running the latest version of Urpl-E.
60. Good enough "in the style of" can create new sales!!! https://twitter.com/Giovanni_Lido/status/1617490801373896712/photo/3
That one guy looks like Mr. Bean though.
then Bill Murray, Alec Baldwin and Christopher Guest (clockwise), right?
The Blackadder of Cranach the Elder's time (1470-1550) was beardless.
||
In my circles, the phrase "Only in America" is usually said sardonically about gun violence etc, but I perceive it as taking off on a genre of feelgood propaganda, usually about meritocracy. Does anyone know the context or era whence that commonplace that America is (somehow) the only place where people come up from low origins? Postwar era? Earlier?
Man, I forgot he was American. Embarrassing (mostly because he was a pedophile).
The author of Ragged Dick? Say it isn't so.
83: Here's the ngram on that. 1960 saw the big spike.
...which was adapted from a collection of Harry Golden essays.
82: Ah, good point. I was thinking of the Elizabethan one.
It seems like the obvious solution is creating a standard representation, and all art with a distance less than a certain amount from the vector representation is infringement. I think the conclusion is that math is better at describing complex systems than legal codes. Fortunately, generative models will soon be creating legal briefs too.
@93 similar ideas have been used for credit card transaction fraud
Fortunately, generative models will soon be creating legal briefs too.
Lack of a human creator? I assume the problem would be similar to the monkey selfie copyright dispute
As it turns out, the US Copyright Office has just cited that selfsame monkey selfie case as legal precedent for not giving copyright to AI art. (Issue still in formal dispute.)
The difference between monkeys and AI is that monkeys are more likeable.
Yeah, plus its easy to find discount slurp juice these days.
Except I guess digital monkeys don't have copyright after all, just in case you thought NFTs couldn't get stupider.
99: Wait, that article seems to be founded on a false assumption. I thought in the current era creators/owners get copyright whether or not they took the trouble to register it with the government.
Of course, if the transfer of that IP is part of your business model, I imagine you should have registered it, as a best practice, but even then, I think you could hand it over in theory, NFT shenanigans aside.
Copyright is automatic, but my read of that article is that they're claiming that since the NFTs are algorithmically generated they're not subject to copyright at all. Which also seems weird (someone drew the original ape!) but may just be unsettled legally at this point.
But both the situation and the article are confusing so I may have that wrong.
From a quick skim of the court filing it looks like they're claiming trademark rights, not copyright.
Trademark makes a lot more sense than copyright for what they are, but passing along trademark as part of a digital contract wouldn't really really work in the way that passing along copyright has been claimed to.
Yeah, what it looks like they're saying is that they are claiming trademark, not copyright, and they haven't registered copyright in the NFTs but they're not saying they don't have copyright in them. Just that that isn't what's at issue in this case.
I guess because the other guy brought up the lack of registered copyright as an issue? What they're saying is that it's not relevant.
I have taken
copyrights
that were in
the court house
and which
you were probably
saving
for AI
Forgive me
they were needed
for me
and monkeys
I'M GOING TO SUE YOU!
My monkey's lawyer says parody is fair use.