It's pretty funny that most people that hate AI art and want it banned because of copyright violations, don't understand that the only difference that will make is that instead of having open models, you will only have closed models owned by Adobe who will license all the imagery from Getty, Disney, etc.
AI art isn't going away, and it better be open instead of in full control of just 2 or 3 companies.
Baffling to me they don't see this, attacking stable diffusion will only harm open and free AI tooling, megacorps will always just "unethically" train.
Also find it laughable that a lot of the most vocal artists against it make their money drawing characters they don't own anyway.
Approaching this as a designer so obviously these tools will impact me but as long as I can run them on my own hardware without paying tax to Adobe/OpenAI I can at least benefit.
The AI process isn't going away, but suits like these will preserve the opportunity for artists and coders alike to get paid for their work or choose the terms under which it is shared.
Without enforcing copyright rules on AI, it may mean the end of copyright entirely.
I share a lot of images under CC-BY-SA-NC and share code under the GPL. If someone can train a model off my work and disregard the terms under which I have shared it, copyright is my only enforcement mechanism.
What you mean by "enforcing copyright rules on AI"? Regardless of the tool used if the code used by someone heavily copies (against your license) on your code then it's copyright infringement, the fact that the tool is an AI doesn't really change this fact. But I guess you want to expand the copyright law to consider more than just the final product (code in this case), as it would be original in the vast majority of cases.
The cat is already out of the bag on that one. You can download several open source text to image AIs and run them on your own computer. You can train new styles and concepts on your own computer in about 20 minutes. You can throw up roadblocks for the big players, but it’s far too spread out to ever be contained again.
Ditto for LLMs. Obviously GPT-4 is the best game in town, but I have several LLMs and the code required to use them installed on several machines. Training new data for them can close the gap quite well at the expense of becoming less general.
We can’t regulate our way out of this one. That window closed a while ago.
If I rob a bank then you can tell. If I make an AI model on my own computer and train it, how would you ever know? It won’t have a “signature” that can be detected by any kind of AI detection system because I’m the only person who knows it exists.
You can make it illegal if you like, but you’ll never be able to enforce it. And other countries are not obligated to follow suit, and will have significant competitive advantages as a result.
There are plenty of things that you could do 'on the sly' that are illegal. That's doesn't mean they're not illegal, and that doesn't mean that if you are found out (which is hard but not always impossible) that you will be punished. Given that there are plenty of legal ways of making money, why would you pick an illegal one just because you (think you) can?
The boundaries of the law are rarely set at what you think you can get away with, but are usually set on some other principle and in between that line and the point where you exist there is plenty of room for prosecution. The jails contain lots of people who thought they 'could get away with it'. Don't be one of those.
There's plenty of public domain material out there to train on. The problem is just that having to filter out copyrighted material from the training dataset would be prohibitively expensive.
These AIs do not need to learn from your work in order to function, they just need to learn from some work. All that extending copyright in this manner would do is create a bunch of unnecessary busywork and slow the progress of technological development. You're not going to get paid $5 for letting a proprietary AI train on your drawing, blog post, or GitHub repo when they could get the same benefit from paying someone else $0.05 to find a public domain image, blog post, or repo elsewhere.
Basically, the amount of value your individual work contributes to generative AI is minuscule, but the amount of effort the entire industry would need to expend in order to compensate you for that value is massive. Unless stifling progress is your explicit goal, it's not worth it.
The difference is that the open systems will be for personal use and the closed models by companies using it for profit. And the reason Adobe and others would be used in for profit settings is because they pay the artists. Which is what's intended by the lawsuits?
I suppose you could argue the beneficiaries are the shareholders of Disney, etc rather than its artists. At the same time, maybe this ends with open source systems that artists can host themselves.
Wouldn’t that then make it exactly what detractors want, either a situation where content creators can be compensated for providing input to the models and/or a situation where models aren’t feasible economically at all?
> but having your IP ripped out of your own blog feels like being raped, by open-source software or not
That's most likely an irrational feeling. "Adobe AI" will most likely work just as well without including your IP in their dataset, and be able to output something in your style with the right prompt (and other knobs we'll see in the future) without ever having seen your IP.
Potentially you could even run some kind of CLIP on your image and then feed that into "Adobe AI". And I really see no reason to think something like that would ever be considered copyright infringement.
People have the idea that their IP is very meaningfully influencing these AIs, while in reality their artstyles aren't that unique.
first of all, it is pretty prepotent to deny someone else feelings straight out of the box. i am an artist and if someone would came with the same art-style i do on Blender with an industrial scale speed; with some text input that took 15 seconds to type and MY data is on it, it would feel like a rape.
and i do not think "i am too insignificant to make an impactful change in society" ... if everyone think(ed) like the quote, life probably would be worse?
Do you also hate Human Artists who trained by viewing copies of your artwork and incorporating elements of your style into theirs? Do you sue all Human Artists that create art vaguely similar to yours?
So if a Human Artist did this on an "industrial scale" you would sue them for having similar a style, how do you think that lawsuit would turn out for you? You are actually mad at the moral implications of Capitalism and not the legality of AI art?
> So if a Human Artist did this on an "industrial scale" you would sue them for having similar a style, how do you think that lawsuit would turn out for you?
when it becomes a realistic possibility that a single human can ingest every work ever created, and then output a derived work in a second to thousands of people at a time: I'll let you know
So, reading your comment, I can infer that the main problem is not the fact that the AI was trained on copyrighted data, but the fact that its performance outperforms that of a human artist.
But the question was not so stupid as it underlines the actual problem of AI, the scale on which it operates, not the fact that a machine could learn similarly to a human. I found the question pretty clever.
I'm trying to point out the flaw in your argument not the real possibility of a human doing this. You have unrealistic expectations of AI art and such a lawsuit against a human would obviously not go in your favor. 5 seconds of google will tell you that Style is not copyright protected so your entire premise is flawed.
There's a parallel in policing: driving around and looking up plate numbers is perfectly fine in the opinion of some, but automating the same task by fixing cameras to police cars and using computer vision to do the task makes it qualitatively different. Scale can have an impact on the way tasks are viewed.
Computers aren’t humans. Software isn’t a person. If you’re going to argue as if computers are now people and deserve having laws apply to them, think about the implications of you owning a human.
> Can they even license those works to train a model?
Adobe is already creating such AI with these kinds of partners, so yes
> And if they can, then everyone can. I can buy the rights to a single photograph of Darth Vader and create an entire series TV about his life using AI.
No because a single image of Darth Vader alone won't work. You need a pretrained base model, trained with millions of images. Only then you can feed such base model a single image and get something decent.
> Adobe is already creating such AI with these kinds of partners, so yes
So, nobody has contested them yet. I smell a class-action lawsuit for them from the original makers of their clipart library.
The base model can be trained on open source images. Or people may use adobe's software for that. From just a few images of Darth you can make basically anything, so yes it's quite doable already.
It seems to me there is no way to put this cat back in the bag ever again.
In the case of Darth Vader, there is probably a trademark involved as well, but the vast majority of photos , celebs etc don't have that.
> It seems to me there is no way to put this cat back in the bag ever again
Patents. Copyrights. Franchising?
Artists are now experiencing the joy of getting beaten in the market by someone else selling the product they created. It was not allowed to stand in other industries.
Copyright is a necessary evil. Disney may well be the savior of artists in the age of AI. There is no incentive to produce anything original if someone else can just gank it and drown you in volume.
> Erik Wallace, a computer scientist at the University of California, Berkeley, told me in a phone interview that the lawsuit had “technical inaccuracies” and was “stretching the truth a lot.” Wallace pointed out that Stable Diffusion is only a few gigabytes in size—far too small to contain compressed copies of all or even very many of its training images.
>In reality, Stable Diffusion works by first converting a user’s prompt into a latent representation: a list of numbers summarizing the contents of the image. Just as you can identify a point on the Earth’s surface based on its latitude and longitude, Stable Diffusion characterizes images based on their “coordinates” in the “picture space.” It then converts this latent representation into an image.
Regardless of the lawsuit, this seems like a quite spurious, self-serving argument in and of itself. If the “coordinates” of the image were not equivalent to the image, how would it be possible to fine tune the model on a set of images and recreate their style? This is setting aside the myriad examples where SD regurgitates fairly close copies of individual works.
This is a bit like training a model on the works of Shakespeare, but then claiming that the tensors resulting from ingesting the data are not actually Shakespeare’s writing. Oh really, where’d they come from then? Thin air?
> Regardless of the lawsuit, this seems like a quite spurious, self-serving argument in and of itself. If the “coordinates” of the image were not equivalent to the image, how would it be possible to fine tune the model on a set of images and recreate their style? This is setting aside the myriad examples where SD regurgitates fairly close copies of individual works.
This argument confuses an image and the style. The former is copyrightable, the latter is not.
If trained correctly the model could easily reproduce any arbitrary image if you know its coordinate in the latent space (and you use deterministic sampler), regardless of whether the image is present in the dataset or not, it's also easy to imagine as the coordinates are as big as the samples generated by the diffusion model. You can take an image and use DDIM inversion to find the coordinates.
Sounds like a "color of bits" issue. In practice the courts would care about where the latent coordinates originated from, and if it were only possible to derive them from violation of copyright. Same issue with a copy of Titanic ripped from a decaying VHS and compressed into a bad quality AVI uploaded on a torrent tracker. At some point there's still an issue with unauthorized reproduction even if the bits don't match the master copy, because the "color" would still match. And "color" is not a thing that can be represented with data, it's abstract information.
Yes but an alternate view of the copyright is that a person get a degree in English literature and carefully reads all of Shakespeare's writing, and then goes onto write a screenplay which thematically derives from Hamlet.
Obviously you get the Shakespeare influence, but the bar on copyright infringement is very high for a reason — we don't want to sue artists for being influenced by another writer.
I don’t really care about the copyright argument as it relates to human learning because I think this is all a Napster-style false flag by capital owners to enclose ALL data on the internet. The intent isn’t to argue about these models from first principles such that we all have IP-restriction free access from them. It’s to make them do enough legally gray things at scale that they get restricted to only a few power players.
Rubbish. Make a reasonable argument instead of playing at being a communist.
'Capital owners' bankrolled the company that manufactures the device you used to type your asinine comment, as well as bankrolling/funding the companies that manufacture every single item you own, from your toothbrush to your clothes.
If you despise capitalism so much, put your money where your mouth is and relocate to Cuba.
Champagne pseudo-socialists like you give the rest of us liberals a bad name.
I made a reasonable argument: the intent with Stable Diffusion is to do the same thing that many startups have done, i.e. break the law enough that you establish a dominant market position/raise a ton of money, but create enough externalities that the state has to intervene. Then, bribe politicians to legalize your former law breaking and to lock out any of your competitors. This is what myriad companies in the gig economy have done, for example.
Instead of reflexively spewing a series of ideological non-sequiturs, perhaps engage with what I actually said.
>bribe politicians to legalize your former law breaking and to lock out any of your competitors
Could you name a single startup in the US that has been caught 'bribing' politicians?
And stop drinking the Uber/WeWork/Airbnb/ kool-aid. The vast majority of startups are founded by honest people, and don't rely on breaking laws in order to establish a dominant market position or 'raise a ton of money', I should know, I'm an early employee at a startup with a 'dominant market position'.
Yes you should demonise people for any criminal actions they've actually committed, but please don't demonise them simply for being successful.
All of your comments on this thread sound as though they're made by a college undergrad still in the anti-capitalist/hippy-communist stage.
Grow up! Most things in life are not conspiracies and believing they are will significantly harm your emotional health.
Oh, but of course, only if I set aside the “kool-aid” of citing the examples of the dominant startup players of the past decade.
Perhaps you should, again, grow beyond your ideological name-calling to a rational engagement with the world, which actually involves everything I’ve described.
I'm late to this, but both of you broke the site guidelines badly in this thread. Please don't do flamewars on HN, and especially please avoid tit-for-tat spats, in the future. We ban accounts that get into this sort of thing regularly; it's not what this site is for, and destroys what it is for.
You've posted some good comments in other threads, so hopefully this should be easy to fix if you review https://news.ycombinator.com/newsguidelines.html and take the intended spirit of the site more to heart. We'd be grateful!
I did a word search of the linked article you posted, and I didn't see a single incidence of the words 'bribe' or 'bribery' in it.
Or are you claiming that the word "lobby" in the phrase "attempts to lobby Joe Biden, Olaf Scholz and George Osborne" actually meant 'bribe'?
And despite you frantically scouring Google for something to backup your 'bribery' claims with, the above article ended up being the only thing you could find.
So basically, you essentially spent your own time to prove my point. Thanks.
I'll repeat again what I told you earlier: grow up and stop gorging yourself on conspiracy theories.
I'm late to this, but both of you broke the site guidelines badly in this thread. Please don't do flamewars on HN, and especially please avoid tit-for-tat spats, in the future. We ban accounts that get into this sort of thing regularly; it's not what this site is for, and destroys what it is for.
You've posted some good comments in other threads, so hopefully this should be easy to fix if you review https://news.ycombinator.com/newsguidelines.html and take the intended spirit of the site more to heart. We'd be grateful!
Do I genuinely believe that startups funneling millions of dollars to politicians in campaign donations, speaking fees, and giving them plush gigs when they leave office is a legalized form of bribery? Yes. Is most of this illegal in other developed democracies? Yes.
I feel like after the whole SBF fiasco this should all be quite obvious…
> Regardless of the lawsuit, this seems like a quite spurious, self-serving argument in and of itself. If the “coordinates” of the image were not equivalent to the image ...
It's simply not. Stable Diffusion 1.5 was trained on over two billion images. Even if we assume the images were only 10kb on average, that would require 20TB of storage space. The pruned 1.5 model is only 4.27GB, or 4,684x smaller. It is fundamentally and provably impossible to achieve that sort of compression ratio on diverse and already compressed images.
'Coordinates of the image' is a layman's explanation of latent space and the training process, it should not be taken literally.
> how would it be possible to fine tune the model on a set of images and recreate their style?
The fine tune process represents additional training, and to reliably and consistent reproduce a consistent style generally requires overfitting. It is very roughly analagous to an artist using a reference while drawing/painting.
> This is setting aside the myriad examples where SD regurgitates fairly close copies of individual works.
Some works were overrepresented in the dataset because they are overrepresented in the world. There are substantially more images and variants of the Mona Lisa on the internet than there are Random Twitter Artist's Furry Waifu.
This works in much the same way for humans, who are more far more likely to be influenced by the Mona Lisa than Furry Waifu.
> This is a bit like training a model on the works of Shakespeare, but then claiming that the tensors resulting from ingesting the data are not actually Shakespeare’s writing.
.. They aren't. Shakespeare did not write them. His works can't be reproduced from them. There is, provably, not enough data for it. This is more different than if an individual read and referenced Shakespeare while writing a play, for all that entails. Specifically: whether it's a human or a model generating the content, the content that is generated could be a violation of copyright.
You'll note that this is no different to if I copy and paste Mickey Mouse onto a tshirt design, or if I draw him on a piece of paper with a crayon.
Should we also be up in arms about copy/paste and crayons?
> Oh really, where’d they come from then? Thin air?
This hits on a more important point than I think you realize here. Where did inspiration for the human artists come from? Did they spring from a formless void? Or have they seen a lot of images, internalized what they think was important from it, and then generated art from that?
You're not exactly communicating in good faith here, given that I addressed exactly this:
> The fine tune process represents additional training, and to reliably and consistent reproduce a consistent style generally requires overfitting ...
It's clear you have a perspective that is set in stone, and insist on not understanding the technology. Hopefully you'll re-evaluate when the plaintiffs in these suits inevitably lose.
No, I don’t care about the lawsuit because the intent with releasing SD was to get sued and then lock in a monopoly based on big players being the only ones capable of rights enforcement. See my other comment.
> I don’t really care about the copyright argument as it relates to human learning because I think this is all a Napster-style false flag by capital owners to enclose ALL data on the internet. The intent isn’t to argue about these models from first principles such that we all have IP-restriction free access from them. It’s to make them do enough legally gray things at scale that they get restricted to only a few power players.
It doesn't really matter if you can't put the AI back in the bottle so to speak. If (in the USA) it's found that AI generated images (based on copyrighted materials) are not fair use, then effectively it will be poisonous to use in your (USA-based) company. This will definitely prevent it from being mainstream.
That being said, nothing is stopping Getty images, for example, from releasing their own image generator, trained from images they own copyright to.
The tricky bit is that those stock photos include generative images too. Many of which aren't labeled as generative. So I wonder if theirs is already including potentially problematic content.
I would be surprised if they hadn't externalized that risk when they started their stock photo library.
Generally, whenever a company buys a photo, there's a release form where the claimed-author of the photo has to state that they own the rights to the photo. So, the infringement is on the part of the seller, and you have a high barrier to sue Adobe for that.
> So, the infringement is on the part of the seller, and you have a high barrier to sue Adobe for that.
No, you don’t, Adobe just has a low barrier to suck the source of the image into the case as an additio al defendano and a potential additional de¹1¹fense against any claim of willful infringement.
OTOH, Adobe would like potential plaintiffs to think that it works the way you describe.
It just means there will be a Google Books-style settlement where a few major tech companies write big checks to a few major publishers (whose work wasn't even verifiably used in training the models.) The AI models will exist and be ubiquitous on the Internet, you'll just have to pay a small tax to some company in order to use them in commercial work.
Do they own the right to train-a-model? It's not clear at all. A lot of great photographers don't want their unique style to be multiplied ad infinitum. Copyright is to "copy, distribute, adapt, display, and perform a creative work", and this is not an adaptation
You can't prevent others from using a style though. Which is why there are so many bands that people might remember having their songs appear in car commercials, but it turns out the car company just paid for a sound-alike clone of a band and not the actual band content.
> It doesn't really matter if you can't put the AI back in the bottle so to speak.
Exactly. Companies like Getty, Disney and Shutterstock are never tired of collecting money and the easiest target for this is Stable Diffusion, which is already commercializing the model with DreamStudio which includes Getty's images without their permission.
O̶p̶e̶n̶AI.com avoided this by partnering with shutterstock to get this permission first before training on their images for DALL-E 2. If O̶p̶e̶n̶AI.com was able to do that, then SD has no reason other than taking images without attribution or credit and using them commercially. (Which they have been caught many times)
Short sighted artists claiming that Getty and Adobe models are fine when in 5 years all tools will be piped through these models anyway, in 10 years GPUs will have them built in running real time and the only difference they made with this hissy for is if they pay Getty and Adobe $100+ a month for the privilege instead of it being free.
You can’t escape it, we may as well all benefit instead of this silly crusade making only 4 companies max benefiting, forever.
Shouldn't it be decided on an image-by-image basis? If the resulting image doesn't infringe copyright, what's the relevance of how it was made? That's how it's worked up until this point.
I mean, maybe because the impact is so severe we should HAVE new laws affording more protections to artists, but that's a different argument right?
Copyright holders have been attacking copyright-infringing tools for quite a while now, there's nothing new about that.
Imagine an image-generating ML dataset that has been trained on nothing but images of Mickey Mouse. Surely the Disney corporation would come after that, I have no doubt about that. Why does commingling Mickey with other inputs change that, if the tool can still reproduce images of Mickey Mouse when asked for that?
I downloaded the Llama model using a torrent. That is, I think, enough of a counterexample to show that lawsuits will not stop this. Raw economics are at play here and the legal system usually loses to economic realities.
1) We just need 1 jurisdiction that allows model training.
2) There is no mechanism to stop model distribution short of banning general purpose computing and the internet. That would cause more harm than good.
3) It is already hard and going to get harder to prove these models were trained on any particular piece of art.
I think copyright is a real world counter example to this kind of argument (that raw economics will override the legal protection). If you are selling anything, you have to respect copyright. So if training on copyrighted images becomes illegal, it doesn’t mean that people won’t be able to create stable diffusion models (they will do it regardless). But it does mean that businesses won’t be able to use them, which removes almost all financial benefits from creating them.
Proof both ways is difficult. How will a company prove that an image was generated with an illegally-trained model? Unless there becomes some requirement that models be kept around for a century.
>Building cutting-edge generative AI would require getting licenses from thousands—perhaps even millions—of copyright holders. The process would likely be so slow and expensive that only a handful of large companies could afford to do it. Even then, the resulting models likely wouldn’t be as good. And smaller companies might be locked out of the industry altogether.
Isn't this what Adobe has done? The results are actually really good.
This weekend I've played around in depth with Stable Diffusion with Controlnet, negative prompts, and inpainting.
It didn't take long to convince me that this AI is "learning" art, in some kind of very similar way a Human is.
Controlnet + SD was really mind blowing all around.
I understand why artists are depressed now, but people are already training hypermodels, textual inversions, and loras.
Who knows what the future will bring ?
My two favorite hyper models are around oil paintings, and cars(for my husband, he is getting an AI car calendar for his birthday, let's see if he figures it out :)
I don't think that impacts legal things, however you are correct, AI art can never go away.
If the legal hammer falls on the bad side of this, corporations/people with decent sized assets are unable to use these tools without paying some sort of royalty.
>I understand why artists are depressed now
As someone who isnt an artist, I wish I could sketch something quickly, use controlnet, and have my idea turned into reality in a few seconds. Non artists are basically at the whim of other creators + the randomness of txt2img. Heck, it seems there is something I want changed with nearly every AI art picture, with actual art skills, you could fix these.
Just to be clear on my point, if I could sketch something, I could probably have perfect AI art in 10-15 minutes. Without those skills, it takes ~1-2 hours.
This is a pretty good summary of the complexity of the issue legally (worth reading the whole thing). The plaintiffs pretty clearly either don’t know how ML works or are deliberately misconstruing the facts. But the real question is how a judge and/or jury will look at this given how tricky the debate is (as evidenced by comments in the many posts about GPT-like models right here on HN).
There are a lot of emotions around this space right now. I’m very curious to see how the legal teams break things down logically and with precedents.
I see why you're saying what you are about the "facts," but I'm not at all convinced that it's so clear-cut that the plaintiffs are all that wrong.
Which is to say, compare these models to e.g. simple image compression. If compressed images can violate, the argument that "this is merely VERY GOOD and VERSATILE image compression" probably has enough legal legs to do something with.
It is a bit like saying a random number generator is a form of image compression since it is a small program the output of which contains every copyrightable image.
Compression has the property that almost all inputs are reconstructable. Generative models have the property that almost all inputs are not reconstructable.
Compression with extra steps is still compression. Most people don't understand the math involved in LZF-encoding, and lossy audio encoding seems even more strange, so "diffusion" just sounds like a somewhat-lossier form of compression if described in a certain way.
I was very dismissive of the anti-AI-image crowd until I saw that Anne Graham Lotz pic. Only a single known source image of her exists, although many copies of it are distributed far and wide across the internet, and using her name in a prompt results in an image that seems very much like a lossy reproduction of the "original." It's an edge case, to be sure, but that demonstration looks like lossy compression, even though it's not.
> I was very dismissive of the anti-AI-image crowd until I saw that Anne Graham Lotz pic. Only a single known source image of her exists, although many copies of it are distributed far and wide across the internet
Also, besides the 20 or so copies of this specific pose I see on the first page, I also see numerous others of her either in a different pose or in a different (but similar) outfit -- she's apparently fond of that color combination.
I just put my local copy of Stable Diffusion to work cranking out images based on an "Anne Graham Lotz" prompt. So far, none of them look anything like the original image. It has generated one of her wearing a similar-colored shirt, but the pose is completely different, and as noted above, she seems to like that color.
From the article: researchers were able to reproduce training images (and bad versions of them at that) 0.03 percent of the time. Not 3 percent, mind you, but 109 out of 350,000 images could be reproduced. And if you go back to the paper, these tended to be images which were heavily duplicated in the training set, so the failure modes should be easy to guard against...
Sure, but I strongly suspect that in court the visceral impact of "Look, there's the image, it was in there," of even just ONE might just be enough to sway a judge or jury, percentages be darned.
Eh, there's a lot of space for argument and deliberation in a courtroom, moreso than a hacker news comment section... This is a rare issue with an obvious fix, so I wouldn't expect a ruling to hinge on it. (Additionally, the Lott image doesn't have any real harm; it's heavily reproduced already and freely enough licensed to appear on Wikipedia.)
My guess is that because there's a lot of these cases coming the courts are really going to want to come up with some tests for whether a model is ok or not. So even if they say SD did a bad thing with these few images, that's not going to be enough to rule out the broader class of algorithms - next week they'll have another Getty case against the algorithm with the technical mitigation for the rare copied images.
But a "rare issue with an obvious fix" sounds like a good Streisand effect type-target to me.
Technologically I know that this is a pretty honest statement of the problem, but I'd expect A LOT OF PEOPLE to read this as "oh, look, they're trying to tweak it to 'destroy evidence.'"
The lawsuit would simply hand over the AI art market to Adobe who trained their AI using their stock photos. Same thing elsewhere - huge companies can pay just enough people (and intimidate the rest) to keep legal.
Fair use would work if we could find an artist for a given generated picture - but we can't. Generate a picture and there is no artist for it closer than the ones who made the pictures in the training set.
I really hope these lawyers wont mess that up, it's an easy win.
I believe that a further restriction of fair use to what is produced by a generative AI would also give a lot of tools for large copyright holders (e.g. Disney) for suing and removing fan art for anything that "is in the style of" or "makes use of the same characters" including uses that would previously be considered beneath their notice and thus not worth their time (the amount the lawyers would cost combined with the difficulty in doing it).
The lawyers won't "mess it up" but rather that the large copyright holders will be able to use this to enforce these restrictions on human generated fan art just as easily as they (and other humans) will be able to enforce it on generative AI.
While it won't be possible to excise the idea of "draw like Disney" from someone's brain, it will be possible to sue the creator to remove the work.
Publishing an image always has a human and has always been the issue.
Doodles, personal art notebooks, photoshopped images - those have never been things that have run afoul of copyright... until they are published.
A person doodling on some paper or running stable diffusion and using it as a background image for their desktop - these have never been at issue before and from a legal point of view are likely very much the same.
I mean, a big problem is that "fair use" is pretty broken already. It's just a really sloppy legal test that ends up bowing to trends, non-expert opinions and utter randomness as much as it does legal rigor.
"legal earthquake" Why is everyone talking in apocalyptic terms about AI? A new legal framework will have to be constructed around this, just like with any new technology. Copyright holders will get relief or we don't have a legal system that works. Have we lost faith in our system so much that we think we live in a disaster movie?
In France, an artist owns its art for life, even if it’s sold.
So let’s say an artist creates a new style, like Picasso did and an other artist creates art that is really close to Picasso. He will lose in court because of IP.
Now an AI does the same thing and suddenly no copyright ?
What should be done maybe is explicitly asking for every art generated to ask to pay for it and this money goes for the copyrights.
But unlimited free art without restrictions nor credit to the original author is stealing IP.
Or you remove everything with IP from the training set because it’s cheating.
Moral rights are... at best a mess. They can be very wide, very up to interpretation and vary depending on the country. They're essentially meant as the ultimate stop button for an artist if their works get used in a way the artist deeply disapproves of[0]. Actual interpretation can vary a lot and cloning an artists style as specifically as SD allows you to do (with "...in the style of X" prompts) could easily be an easy moral rights argument in most jurisdictions, doubly so if economic exploitation is a component. (Which a lot of the hype surrounding SD et al. is also build on.)
For this reason they mostly exist in civil law systems (where a judges ruling sets no precedent for future interpretation), while they barely exist in common law systems (like the US) since it's really difficult to point at consistency with previous cases when it comes to moral rights.
Largely irrelevant - like I said, moral rights exist in civil law countries, so these things will be pursued on a case by case basis rather than by existing precedent. They tend to deal with (quoting WP here) "The preserving of the integrity of the work allows the author to object to alteration, distortion, or mutilation of the work that is "prejudicial to the author's honor or reputation".[3] Anything else that may detract from the artist's relationship with the work even after it leaves the artist's possession or ownership may bring these moral rights into play." Another condition often applied is straight up forbidding the modification of a copyrighted work, regardless of any other circumstances.
In this case, you can reasonably consider that the strong stance that a good chunk of artists have taken about AI art, that using their work in these models is deliberately altering/distorting the work (Since training these models involves serializing the art down to patterns that can be replicated and iterated on) in a way that is damaging to their reputation.
As for whether or not relief has been granted, do keep in mind that requiring relief for a lawsuit is largely a US only thing. The court systems of other countries tend to lean more towards stop orders rather than immediately pulling out the financial compensation card. That's also the context in which moral rights exist - they are intended to be a stop order, not intended to be something you can financially gain from having it be violated (which is why they exist separate from economic rights, which are well... typical copyright stuff).
Do keep in mind that I am not a lawyer, US or otherwise.
> As for whether or not relief has been granted, do keep in mind that requiring relief for a lawsuit is largely a US only thing. The court systems of other countries tend to lean more towards stop orders rather than immediately pulling out the financial compensation card.
A issuing a ”stop order” (e.g., in US law, a permanent injunction or something similar) is granting relief, financial compensation isn’t the only kind fo relief.
> Le droit moral de l’artiste est «perpétuel, inaliénable et imprescriptible». Il est lié à la personne de l’auteur qui ne peut y renoncer ni le céder à autrui
> The moral right of the artist is "perpetual, inalienable and imprescriptible". It is linked to the person of the author who can neither renounce it nor transfer it to another person
> Le code de la propriété intellectuelle entend par contrefaçon tous les actes d'utilisation non autorisée de l'œuvre. En cas de reprise partielle de cette dernière, elle s'apprécie en fonction des ressemblances entre les œuvres. La simple tentative n'est pas punissable.
La loi incrimine au titre du délit de contrefaçon :
- «toute reproduction, représentation ou diffusion, par quelque moyen que ce soit, d'une œuvre de l'esprit en violation des droits de l'auteur, tels qu'ils sont définis et réglementés par la loi » (CPI, art. L. 335-3).
- «le débit [acte de diffusion, notamment par vente, de marchandises contrefaisantes], l'exportation et l'importation des ouvrages "contrefaisants" » (CPI, art. L. 335.2 al. 3).
> The Intellectual Property Code defines infringement as all acts of unauthorized use of a work. In the case of partial use of the work, it is assessed according to the similarities between the works. The simple attempt is not punishable.
The law incriminates under the offence of counterfeiting:
- "any reproduction, representation or dissemination, by any means whatsoever, of a work of the mind in violation of the author's rights, as defined and regulated by law" (CPI, art. L. 335-3).
- "the debit [act of distribution, notably by sale, of infringing goods], the export and import of 'infringing' works" (CPI, art. L. 335.2 al. 3).
I’m not an artist so I have no problem with being able to generate almost free quality art on my own within minutes.
Before those AI if I ever wanted something specific I had to go see an artist and ask (prompt) him for the art I want. If I ask him to draw Andy Warhol Marilyn with my face on it. Can he charge me for it without copyright infringement ? Maybe nobody cares and no one is going to do anything about it but still, his style is specific right ? My face or my dog’s face makes the drawing completely different.
If AI is just « an other form of intelligence », why shouldn’t it be subjected to the same laws ?
If it’s just a tool and not an intelligence, why would openAI wouldn’t be subjected to the same rules ?
If all you do for a living is painting, if I wan’t a unique painting that fit your work style I have to ask you to do it.
Now I can ask an AI to paint exactly that but you are not in the loop anymore but your work has been used to train this AI and no one will ever ask you to create art because they love your style and can have it for free.
I mean, I’m all in for free stuff and I’m not an artist but what does artists feel about it ?
> If I ask him to draw Andy Warhol Marilyn with my face on it. Can he charge me for it without copyright infringement ? Maybe nobody cares and no one is going to do anything about it but still, his style is specific right ? My face or my dog’s face makes the drawing completely different.
You can generate whatever you want. If you want to sell it, then maybe it is a copyright violation. Maybe it isn’t. The point is that it does not matter that an AI tool made it or a human or a random falling of colored sand. Nobody is saying AI tool outputs are immune to copyright. What I am saying is that they are not intrinsically violations of copyright because training data.
> If all you do for a living is painting, if I wan’t a unique painting that fit your work style I have to ask you to do it. Now I can ask an AI to paint exactly that but you are not in the loop anymore but your work has been used to train this AI and no one will ever ask you to create art because they love your style and can have it for free.
That is deeply misleading. You are absolutely fine to ask another artist to create you a work in another artist’s style. The only difference here is cost and perhaps quality. Those are not problems with copyright.
> I mean, I’m all in for free stuff and I’m not an artist but what does artists feel about it
I do agree that AI is a great tool to generate whatever we want and I have no problem with that.
My only concern is abuse in the case of copyrights. And a partial copy is still under copyright and IP. Because the services aren’t completely free they might be subject to IP.
Let’s say I generate a picture with the style of a famous artist and then use it in one of my marketing campaign. No one will be able to sue me ? You are 100% sure about it ?
If so, then then could also sue the AI company also isn’t it ?
It seems like a total abolition of copyright otherwise
So you can be charged if you create a painting in the style of Picasso and try to sell it as a true Picasso work. Which is obvious. Nothing here says Picasso's style in of and itself can be copyrighted.
Which came first, copyright or the printing press? Was the printing press banned because of copyright?
But yes, the claim about an artist owning an entire style sounds pretty off to me. There's no way Picasso has a copyright claim on every cubist painting ever made, nor does Van Gogh own every painting containing a sunflower.
> Which came first, copyright or the printing press? Was the printing press banned because of copyright?
I feel like you’re making the same argument as me, with the same rhetoric, but a different example. No, I do not think the printing press, paint, or generative tools are bannable by copyright. I do think all 3 could be used to produce a work that is in violation of copyright. And no, I don’t think a novel work produced by any of them are in violation of copyright.
I think it's just not clear what argument you're actually making. When you say "a novel work produced by any of them", are you equating the works produced by paint with the works produced by AI? Are you saying that if I throw a can of white paint against a Jackson Pollock painting, the paint itself is producing a novel work? That if a million printing presses print out a million dictionaries, one of them is bound to produce the entire works of Shakespeare?
My point is that copyright does not care about the means by which a work is generated. It is merely whether the output is substantially similar.
If a printing press randomly produces Hamlet, you cannot sell it. That does not mean you cannot generate something that is sellable without violating copyright by printing random words.
There is a small amount of attention paid to whether an artist knew of a work that someone claims was copied. Especially in the literary world with hacks who try to steal from successful authors with pathetic claims.
But they tend not to hold a lot of weight. It just makes it easier to dismiss trolls.
As someone that’s undecided on this whole issue, your particular argument comes across as so tone-deaf that I didn’t think that you genuinely believed it when you wrote it.
Which piece was tone deaf? That AI is a tool? Because it is a tool. Using it to copy a work is a problem with copyright. Using it because it is capable of violating copyright is not.
Or that style is not copyrighted? Because it’s not.
Humans regurgitate other people's work all the time, and these image generators are well capable of creating non-infringing images. How is AI regurgitating other people's work any different?
Basically there needs to be a copyright-court-like oracle used in training, so as to suppress output that's quoting the sources to much.
"“Stability AI has copied more than 12 million photographs from Getty Images’ collection, along with the associated captions and metadata, without permission from or compensation to Getty Images,” Getty wrote in its lawsuit."
The hypocrisy of Getty Image is insane here, given it's dark past.
"Napster copyright lawsuits could be a legal earthquake for peer-to-peer file sharing"
I think the only effective way of slowing down AI isn't copyright enforcement, but trade secrets. If Midjourney can control who has access to its service and what kinds of content its users can generate, democratized models/piracy can do nothing to change that.
The reason why there's so much more interest in AI art than AI voice cloning is that there's not yet a publicly available AI voice model with good enough performance. When that model is released or leaked there will be a similar panic around the effectiveness of modern AI voice tools, because anyone good or bad can get their hands on them without being forced to go through the company's content filter.
Hmm. Graphic-media artists tend to have much less wealth and capacity to enforce their copyrights when compared to the music or film industries, but Getty has both the funds and the political connections (the Koch brothers own about 19% of Getty [0]) to vigorously defend their copyrights.
Not sure if there's a more specific word for it, but it's basically just a straw man attack: constructing a single "opponent" from disjointed pieces of information. As such, you could also classify it as a hasty generalization or a composition error.
Not everyone on hn shares that opinion. Ai products are great but not when trained using stolen ip. At the moment there is no way to prevent openai’s products from ripping code or web content and then spitting it out without royalties. And that’s quite an issue.
I think this comment would be a better jumping off point for discussion if it were responding to a specific individual and showing curiosity rather than making multiple broad assumptions about the community at large.
Our copyright systems are broken, and have been broken for decades at this point.
I don't think copyright should be a thing, ESPECIALLY at the current durations. It's obvious to me that reform is necessary.
Here's what I propose:
- Reduce the duration of existing copyrights to 20 years.
- Reduce the duration of new copyrights to 5 years (maybe 10).
- Provide some form of remittance for individual copyright holders which depends on the income for their living, if the 20 years is not enough.
- Require that what is copyrighted must be submitted to the Library of Congress. When the copyright expires, it is to be released to the public in full.
- Strengthen fair use protections. Make it clear that critiques, reviews, parodies, etc are covered.
- Improve takedown requests. Punish fraudulent takedowns by forcing the filer to pay any lost income to the recipient. Require the takedown requests to have specific information about where the copyright violation occurred and how it isn't fair use.
The changing the term of protection provided by the Berne Convention or TRIPS would mean that the United States would be leaving those treaty... which has quite a few implications for how other countries would treat copyrighted works within the US.
> TRIPS requires member states to provide strong protection for intellectual property rights. For example, under TRIPS:
> Copyright terms must extend at least 50 years, unless based on the life of the author. (Art. 12 and 14)[8]
> Copyright must be granted automatically, and not based upon any "formality", such as registrations, as specified in the Berne Convention. (Art. 9)
I don't believe anyone has a right to a monopoly on information, even if they created it. I don't believe in the validity of intellectual property in general. Even 5, 10, or 20 years is a compromise.
The Library of Congress is just an example, I'm sure some international agreement could be made for an independent entity.
As for the Berne Convention or TRIPS, I wasn't just talking about the US. The international copyright laws should also be reformed.
Why shouldn't I be the only one with the right to sell that image?
The course of action for "lose the rights in 20 years" isn't "ok, that's fine" but rather "ok, then I won't publish any of them and deal only in direct sales of prints from a gallery."
> Why shouldn't I be the only one with the right to sell that image?
The real question is why should you? Copyright is not a natural right (if it was, it wouldn't require enforcement actions on people who never interacted with the original copy). It's a privilege meant to provide an incentive for artists (the effectiveness of which is questionable, even without taking into consideration the negative side-effects).
> The course of action for "lose the rights in 20 years" isn't "ok, that's fine" but rather "ok, then I won't publish any of them and deal only in direct sales of prints from a gallery."
And that is entirely within your rights. But once your photo is disseminated into the public sphere, it's information that you no longer have control over.
AI art isn't going away, and it better be open instead of in full control of just 2 or 3 companies.