Hacker News Reader

Screw one person? A great offense. Screw lots of people at once? A great innovation.

You have a point, but it's also how art works in general. Most artists draw inspiration from a conglomeration of hundreds/thousands of other artists. If an artist draws inspiration from one and only one artist, they're just a plagiarist.

(Not necessarily saying that clears up any potential legal or ethical issue with generative image models training on artists' work.)

But these aren't artists and this isn't art. This is fast food. This is content for articles and technology for startups to become the middelman for yet another thing.

There will be art coming from these tools at some point but right now it's creatively bankrupt illustrations driven by curiousity and lots of creatively bankrupt people.

Yet. As "real" artists start using these generations to leapfrog their projects, at some point, how is this different from someone studying a style and producing something in that style with that addition of actual "art?"

That's like expecting a chef to take inspiration from a fast food menu. It might happen but it's more likely that the chef already knows what makes fast food taste good and can develop a new menu based on their fundamental knowledge.

The fundamental knowledge they gained by looking at thousands upon thousands of previous bits of information; and, after today, that collection of insights is going to include auto-generated artwork, as well.

It's not any different if there's true craft to it. The person in this article and many of the examples are who I'm referencing as not an artist or making art. I would imagine the person that wrote the article doesn't call himself an artist either.

I'm not arguing that the tools can't be used to create art and I'm not trying to say "real" artists don't use these. What I'm trying to say is the images I've mostly seen look like illustrations and content machines.

Personally, these tools are incredible and are almost magic in a sense.

Edit: To put it another way a writer can still write if his keyboard stops working, a carpenter can still build if his saw breaks, an illustrator can still draw if their tablet dies.

Why do you think artists draw inspiration from other artists?

An artist who's never seen art is like an AI with no training data

And based on what I've seen in the article, the AI images look to be notionally better than the original examples.

We probably have different definitions for what constitutes an artist. In my experience most artists spend time looking at the world around them rather than looking at other artists' work.

not my college of arts experience.

All art is derivative. Even "outsider art", which is made by artists who are ostensibly naive to the work of other artists. (Personally I think that category is bullshit gatekeeping, but that's another discussion entirely.)

Great innovations do tend to "screw lots of people". Cars put most buggy makers out of business. Light bulbs, candlemakers. The computer, human computers. The internet, journalists.

AI promises to be a particularly disruptive innovation, but I don't think anyone can or will stop it. Instead, we should think about improving society so that the promise of AI benefits everyone and not simply a select few.

> Anyway, just my personal opinion. It's become a very 'us vs them', lines-in-the-sand argument these days, and it'd be great if the conversation could be less heated and more philosophical.

It's heated and less philosophical because many artists are worried about their livelihood while a multi-billion dollar company is working towards making them obsolete often using their own work.

I don't understand the confusion people have towards this issue.

> many artists are worried about their livelihood while a multi-billion dollar company is working towards making them obsolete often using their own work.

You do realize that most commercial art is "art for hire" and in this very story some of the examples trained were not owned by the artist.

Multi-billion dollar companies already do this. Hire artist to draw Corporate IP. Corp owns and can do whatever they want with it. Maybe they hire that artist again. Maybe they hire someone else and they share the work in a reference folder.

But now they (maybe) get to skip the "hire arist" step. Which, from the point-of-view of the artists, is the most important part.

The distinction I think is that the multi-billion dollar company working towards making artists obsolete by using their own work didn't pay any of these artists for that IP. At least with hiring an artist to draw corporate IP, an artist has to relinquish their rights to that work explicitly and are paid for those rights.

Are we talking about Stable Diffusion or GitHub Copilot here?

I’m not convinced there’s a difference.

> It's heated and less philosophical because many artists are worried about their livelihood while a multi-billion dollar company is working towards making them obsolete often using their own work.

IN that case, odds are we should be outraged about the existence of 99% of people here, because they work for tech companies whose sole goal is to create software that tries to make human employees redundant.

It doesn't really matter: the AI model is a compressed representation of copyrighted works processed by a generic, non-artistic algorithm specifically engineered to extract the artistic features of such works.

Conceptually, it's very similar to running lossy JPEG over a single copyrighted work versus a giant folder of works from different artists, then shipping that compressed collection to you customers so they cut and paste sections of those works in their collages. The output your customers create with this tool might be sufficiently original to warant copyright protection and fair use, but your algorithmic distribution of the original works (the model) is a clear copyright violation, it doesn't matter if it affects a single artist or thousands.

How do you copyright a style though?

> your algorithmic distribution of the original works (the model) is a clear copyright violation

Depends on if it's considered "transformative" enough. If Google can cache thumbnails for search, how is an AI model not a "search database+algorithm?"

That said, I do expect that SD will be shut down for copyright infringement at some point because you are right, the model does have a bunch of copyrighted material in it and Disney's lawyers will probably come swifter and more prepared than the defense.

As a general rule, the fair use thumbnails enjoy is very limited, only in certain jurisdictions, and only for very specific use-cases.

An universal art-production machine that can compete in the market place with the original artist - and indeed crush them on productivity and price - certainly does not qualify as fair use.

>AI model is a compressed representation of copyrighted works

But it's not because you can't get a copy of the copyrighted works back out.

Misconception? I mean, if it talks like one and quacks like a collage machine, that's kinda what it is. It's just using an infinite (well, 4 GiB) magazine reel to cut out from.

In the end, you are not going to be paid for drawing everything all over again. You are going to get paid for devising a unique style to match the creative vision of the movie, series, book or whatever and for tweaking the machine outputs.

There is a chance that style will become copyright-able. Well, eventually common style banks will appear.

I think painters also copy each other styles , or make copy of popular works, artists don't like this but I did not seen people demanding "do not use this art style because XYZ created it",

For me, just a regular person, the art is not something an soulless machine can generate or a monkey with a camera, the intent and mind is important. So a guy can ask an AI to create a malformed portrait of some subject with some artist style , the value of the art is in the subject,theme and not the style IMO. Like if you ask for a portrait of X steping on Y dead body, dressed in Z and with M,N,P the artistic value is in your idea behind this and not in the pixels.

I remember similar complaints when digital art started to get popular, that is not real art , that you just move pixels around

> Feels like there's a difference between artists as drops in a vast ocean of training data, vs explicitly creating a model on one person's work.

I don't.

Stability.ai did the big training sets and it is coincidence that it remembers the names and categories accurately.

If you want to leverage this tool more fine tuned then you add these modules with more accurate naming.

I would be for some way to compensate artists, like if the use of a module gave them a royalty, but I don't think it is an ethical, legal, or social norm to enforce. If it happens in a uncircumventable way, I would be for it. If it doesn't, I'm for that too.

I've seen that with good models (I think it was StableDiffusion, not sure), you could get style imitation of a specific person's style; even if it was using a ocean of works, the model was still successfully fishing the exact drops of a specific artist (or close enough that I couldn't tell the difference). And thus the model was able to rip-off dozen of different styles.

Hey, we want to use machines to steal people's creative work, take away their jobs in the future, and create more algorithm generated garbage since it worked so well for news sites and promoting videos/posts/social media. We also want to pretend that computer generated graphics are created by an 'artist' for the sole purpose of being able to assign copyright to our machine works, but be able to ignore artists and copyright in every other way.

Why do people seem to have a strong opinion against this? It's just stealing the most intimate thing humans create (art) so that we can create soulless algorithmic versions of it for our own uses because evil artists won't let us just do what we want with their works and won't let us profit on their work. Artists are jerks.

> We also want to pretend that computer generated graphics are created by an 'artist' for the sole purpose of being able to assign copyright to our machine works

The current IP rights system allows for work for hire, assignments. Reframing it to "I commissioned an AI" obliterates any debate about who owns what. When you commission the rights of what is created is assigned to you, for this machine model, just add that clarity in the TOS and its a done deal.

Creating art similar to another artist's work isn't stealing when humans do it, why would it be stealing when machines do it?

(Unless the result is too similar, like the Wonder Woman image generated by Stable Diffusion shown in the article, in which case it's "stealing" whether created by human or machine.)

Because machines are not humans, and shouldn't enjoy the benefits of fair use.

Copyright law exists to balance the interests of the people that make up the society, not some abstract caricatural embodiment in algorithmic form.

Should humans using machines enjoy the benefit of fair use?

Sure, if you build your own model, train it on copyrighted works, then use it to create art; or if you use someone else's model which properly license its copyrighted sources, and use that to create art. In both cases your output is a new creative work sufficiently different from its parents to not constitute infringement and enjoy its own copyright protection.

However, the model creator/distributor will never be able to claim fair use on the model itself, which is choke full of unlicensed material and can only exist if trained on such material. It's not really a subtle or particularly difficult legal distinction, in traditional terms it's like an artistic collage (model output) vs a database of copyrighted works (trained model).

The trained model is not a sufficiently different work that stands on its own, in fact it is just a compressed algorithmic representation of the works used to train it, legally speaking it is those works.

In what way is the model chock full of unlicensed material? It was trained on unlicensed material, but I don't think you're ever going to be able to find a forensic auditor who can tease individual works out of the weights in a model.

You can't reasonably assert that a model encodes individual works of copyrighted material in any way meaningful for copyright. Not without a change to the law.

Obfuscation is not a valid defense against copyright infringement. If my database contains full encrypted copies of unlicensed works and I distribute keys to my customers for parts of those works, no forensic auditor will ever prove the full extent of my infringement without learning the full keyset. But I would argue that reproduction of even a single instance of an unlicensed non trivial fragment of a copyrighted work would taint that entire database.

In the same way in AI a crafted prompt that creates striking similarities to a well known work, like this example here, is suficient proof that the model embeds unlicensed works; using a copyrighted work for training models is just another form of commercial exploitation that the original author should be compensated for.

But we're not talking about obfuscation; we're talking about the data you're describing not being there. If you ask the AI to spit out a 1-for-1 copy of Hollie Mengert's work, it can't. I suspect it can't spit out coherent individual pieces of it either (I might be wrong in that assertion, as I haven't run this Stable Diffusion). It spits out content in her style.

You generally cannot copyright a style.

If it spits out entire chunks of pre-existing works, that's an entirely different story; but what it seems to do is (via the learning and subsequent diffusion process) receive an input like "Wonder Woman on a hill" and (to falsely anthropomorphize a giant math puzzle) say "I know what a Wonder Woman looks like, and I know that 'correct pictures' have some certain ratios of straight lines and angles and tend to use some particular color triplets, so I'm biasing the thing that matches to my 'Wonder Woman' shape structure with those lines, angles, and colors." The result is a picture Hollie Mengert has never drawn, which an observer could assume is done by her because the style is so spot-on.

And aping an artist's style is not illegal for humans and we have no law to make it illegal for machines. Should it be illegal is an interesting question, but it will require new law to make it so.

I'm not claiming the problem is identical to pre-existing problems in the copyright space, just that it's sufficiently similar not to pose a significant challenge for legal scholars, IMHO. Existing copyright laws not only forbid verbatim reproduction, but require derivative works too not prejudice the original author, and grants those authors the power to authorize and reject derivation: https://en.wikipedia.org/wiki/Derivative_work

You anthropomorphic analogy falls flat in its face because the algorithm does not "know" anything, not in any sense of the "know" word for sentient and rational creatures. The algorithm embeds an association between the text "Wonder woman" and actual artistic representations of Wonder woman included in the prior art it is trained on. When prompted, it can reproduce one (see the Copilot fail where it spited out verbatim copyrighted code including comments) or a remix of such representations and integrate them into the output. That's plain as day a derivative work.

The particular case you are referring too, style extraction, could be considered fair use assuming you can technically separate the base visual model from the output style and you can prove the training data for the output module is distilled into abstract, statistical quantities pertaining to that style, such as color palete, stroke weight etc. That sounds like a tall order and I would consider any AI-model trained with copyrighted works as tainted until that burden of proof is satisfied.

Isn't the fact that it can faithfully simulate, in the style of the author, works the author has never created proof enough that the style is disjoint from the trained content?

Hollie Mengert never rendered the streetscape in the article, but DreamBooth did it in her style.

If we're talking criminal copyright infringement, why is the burden of proof on the defendant to show statistical abstraction if the plaintiff can't prove the AI generates works she has made? (Again, if it is possible to get DreamBooth to kick out Hollie's original work, or substantial portions of it, I'd be inclined to agree with your way of thinking, but I haven't seen that yet).

> embeds an association between the text "Wonder woman" and actual artistic representations of Wonder woman included in the prior art it is trained on

Not if I understand how it works correctly, no; it does not. In fact, Mengert's rendering of Wonder Woman differs from the one DreamBooth kicked out if you look up the work she's done for "Winner Takes All! (DC Super Hero Girls)". This is because DreamBooth's approach is to retrain Stable Diffusion with new information but preserve the old; since Stable Diffusion already had an encoding of what Wonder Woman looked like from a mélange of sources, its resulting rendering is neither Mengert's nor the other sources, but a synthesis of them all.

I can't quite follow what you're saying due to your choice of subjects. "It's just stealing the most important thing humans create so we can create soulless algorithmic versions of it for our own uses..." Who is the 'we' there if not 'humans?' I can tell you the machines do not care one way or the other about art.

What we're looking at is the ability for humans who have not trained to be artists being able to create something similar to trained output. That's what technology does: augment human capacity to do something. It's what it always does.