Midjourney and Order of the Stick

[Punoqllads]

New court filing by Midjourney in an ongoing lawsuit says that they sampled the artwork of a whole bunch of artists including [drumroll] Rich Burlew. Here's hoping that the resulting decision protects the hard work of all artists from the depredations of our less ethical technologists.

Broken link below because it gives me an error trying to post the actual URL.

https ://storage.courtlistener.com/recap/gov.uscourts.cand.407208/gov.uscourts.cand.407208.129.10.pdf

[KorvinStarmast]

Is this a scraping thing, or something more pernicious?

[Synesthesy]

AI is the genie that come out of the bottle. You can't put him back.

By the way, I use a lot of AI for work nowaday. It fill the gap between a very mediocre artist with good ideas(me) and someone who can produce a product that can be sold. Does that make me able to replicate a work like the Order of the Stick? Oh, no. Totally no.
It is an incredible fast tool that makes people with low budget and some unrelated skill (for example, coding) to be more 'complete' and then able to deliver a more complete product, and that will help making awesome things in already high budget production (like in movies). It will not stole the work from people like Rich any time soon, even if I'm able to ask a software to draw an ugly uninspired Elan-like character different every time.

BUT if there is a chance that some big AI companies need to pay money to people like Rich, I totally support this.

[Punoqllads]

Is this a scraping thing, or something more pernicious?
Stable Diffusion scraped five billion images according to the lawsuit, and Midjourney used some or all of that.

[KorvinStarmast]

Stable Diffusion scraped five billion images according to the lawsuit, and Midjourney used some or all of that. Thank you, I won't be able to read the pdf until later (when I have time to myself).

[hroþila]

It will not stole the work from people like Rich any time soon
It literally already did. All AI models are trained on stolen work. It's procedurally generated plagiarism

[Punoqllads]

It's complicated. You can't copyright an art style.

If a human artist looked over Rich's art and produced some images in the style of OotS, it would be extremely cringe but not, legally, theft. It would not be something that Rich could simply stop via the court system.

Still, it feels wrong for artists to spend years if not decades building a portfolio on the Internet in order to make a living only for some well-funded tech company to yoink their style once they get big enough. If allowed to persist it would ruin Internet art.

[Errorname]

AI is the genie that come out of the bottle. You can't put him back.

It's still worth trying. It's never going to be how it was, but this sort of mass scraping without any credit or consent or compensation shouldn't be allowed.

[hroþila]

It's complicated. You can't copyright an art style.

If a human artist looked over Rich's art and produced some images in the style of OotS, it would be extremely cringe but not, legally, theft. It would not be something that Rich could simply stop via the court system.
It's not comparable to an artist imitating another artist's style though. It quite literally uses the source material directly

[Lord Torath]

AI is the genie that come out of the bottle. You can't put him back.Maybe not. But we can still slap some manacles on it and tell it there are limits to what it's allowed to do. ("I can't make anybody dead. So don't ask.") And we can tell its handlers what they can and cannot train it on.

[KorvinStarmast]

For some reason, this all reminds me of Napster. :smallconfused:

[Synesthesy]

It literally already did. All AI models are trained on stolen work. It's procedurally generated plagiarism

I'm afraid I said it wrong: I mean JOBS, not WORKS; in Italian we use the same word for that, sorry.

Artists' jobs are safe and they will for a very long time. It will open more jobs for mediocre artists like myself. But I don't think that you can stop a software from "watching" other people's works. I don't know. I don't think so.

Am I more clear now?


Still I approve if they will make people like Rich get some money out of it. I don't want AI to be closed, but I totally agree that artists need more compensation. If possible.

[Kardwill]

Artists' jobs are safe and they will for a very long time.

I don't think they are. If AI can do an artist's work quicker and for cheaper, then jobs that would have been handled by low-level artists will be handled by AI instead. Just like cameras pretty much instantly killed 90% of the portrait artist profession.
We are beginning to see boardgames that use AI to replace their artist contractors, for example, and I guess RPGs will also meet the same problem.

Of course, many uses of AI right now are from people/companies that would have been too cheap to hire an artist anyway, so those are not really competition for human artists. But companies will seize any opportunity to reduce their manpower costs, and that means real artist commissions will get outsourced to AIs, and AI artists, that can work much faster (and thus much cheaper) than classic/digital artists.

I agree that the genie is out of the bottle. Any technology that allow profit WILL be used, whatever the consequences. But I think it's unwise to underestimate those consequences. The AI isn't able to take over talented artists jobs yet. But it will compete with low level artists, that will struggle to find paid jobs when they begin their carreer, and thus won't be able to develop into talented/expermimented/famous artists "safe" from the AI.

[Precure]

AI, as we understand now, can only imitate the artist and can never replace the humanity's spark of creativity.


https://i.ibb.co/XYyd87z/1680183373672.png

[OvisCaedo]

Companies absolutely will use AI to replace human artists as much as possible. It's already being experimented with. And while it's certainly not perfect enough to replace humans completely yet, you're fooling yourself badly if you think it isn't the goal being worked towards.

WIT Studio, a Japanese animation studio, put out an experimental short with Netflix using AI to work on the backgrounds, claiming they're looking to help with the "labor shortage" the animation industry has. Though any such shortage is purely because animators are paid and treated horribly, and the goal is to pay even less.

The same studio was recently announced to be remaking the anime of One Piece, once again in collaboration with Netflix, and with the pitch mentioning it would use "cutting-edge visual technology", which has yet to be clarified but is almost certainly going to mean AI image generation. One of the biggest, best selling properties alive today, looking to use AI to cut costs. Though, of course, that is only an assumption right now; maybe "cutting-edge visual technology" will turn out to have been a buzzword for something else.

[Metastachydium]

[IMG]https://i.ibb.co/XYyd87z/1680183373672.png[IMG]

Unholy clouds, WHAT IS THAT?!

[Precure]

Unholy clouds, WHAT IS THAT?!

A supposed scene explaining what Eugene done here:

https://www.giantitp.com/comics/oots0292.html

[Metastachydium]

A supposed scene explaining what Eugene done here:

https://www.giantitp.com/comics/oots0292.html

I find that lack of faithis disturbing.

[Punoqllads]

It's not comparable to an artist imitating another artist's style though. It quite literally uses the source material directly
"Directly" is not correct. Very, very indirectly, in fact.

AI doesn't store the images. It uses something called a neural network. In the network are a bunch of nodes ("neurons") each with a set of numbers ("weights") associated with it and each node connects to one or more other nodes, some of which are upstream and others of which are downstream.

All of these image generation AIs came about because computer scientists were originally building image classification engines. After training, you give as input a never-before-seen image and the engine comes up with a description for it, e.g., "An Impressionist-style painting of water lilies in a reflecting pond, reminiscent of a 19th Century French garden". Someone had the bright idea of running the engine in reverse -- swapping upstream and downstream -- and lo and behold when they typed in a description they got an image out that wasn't half bad.

After a few thousand training images -- much less five billion -- it's pretty much impossible to say where, if any place, a given image from the source data (or the words used to describe it) "lives". For an extra layer of complexity, sometimes they use the network to generate a very large set of data, erase the weights, and then use the generated data to create a brand new weight set.

[Tass]

It's not comparable to an artist imitating another artist's style though. It quite literally uses the source material directly

In what way is it more direct than what a human brain would do?

[Kaytara]

It's still worth trying. It's never going to be how it was, but this sort of mass scraping without any credit or consent or compensation shouldn't be allowed.

I find it extremely odd that people with this position continuously imply that if these AIs were made WITH credit and/or compensation, then it would be a-OK. For one thing, if the same technology had been made with the up-front cost of, say, buying up a large database of assets or paying one dollar per artwork to every person whose art is used - costs that are substantial but not exactly insurmountable for a corporation worth billions making a product that is expected to be very profitable - then the way AI art ostensibly threatens artist jobs and human creativity in general would still be a factor, and that's the thing people are actually mad about. For another thing, the creative industry is rife with examples of corporations interacting with individuals in a way that is technically legal and above-board, but still exploitative.

Personally, I find it good and cool and awesome that the technology exists, and this technology would probably not be possible without mass-scraping, with permission or otherwise. The only problems with this technology are those that predate it - namely, economic precarity in general, and to a lesser extent the flawed incentive structure of everything from the art industry to online galleries. Instagram is a content mill full of multi-million follower accounts reposting art they didn't make, and Facebook - sorry, "Meta" - does nothing about this because it's profitable for them, they don't actually give a single **** about artists as individuals. The site was already unusable before AI art, people are just paying more attention now. Just as one example.

I say "ostensibly" and there's the "it will automate jobs away" angle, sure. But there are no problems with AI art that aren't also inherent to (say) the coffee machine, or the self-checkout machine, or nuclear technology. "Yes this tech is cool but how will it be USED in the current system? How does the inevitable application reflect on the morality of the technology itself?" Not a bad question, but not a new one, either, and it's already been done to death.

Hollywood is full to the brim of cheap VFX because VFX artists aren't unionised (https://www.theguardian.com/us-news/2022/sep/09/visual-effects-artists-union-push-movies) and studios can outsource their VFX labour to what are essentially sweatshops. (https://www.telegraph.co.uk/films/0/without-us-no-film-industry-inside-special-effects-sweatshop/) When the art industry absolutely sucks - for those of us paying attention - and it starts sucking slightly more because of AI art, because those industries now have 10002 tools they can use to exploit their workers instead of a mere 10002 - zeroing in on AI art as some exceptional example of it seems myopic.

The "oh but it's unoriginal!" cry is likewise tiring because since when do we care THAT much about originality? I write fanfiction, FFS. But at least I can share my work freely, unlike this artist who put what must have been like a hundred hours of work into these gorgeous assets (https://www.deviantart.com/rogner5th/art/concept-designs-2-671865978) that went unused but can't even sell prints of it because Ubisoft still owns the copyright! Because yeah, suddenly originality stops mattering when you're being paid for it. Even "better", every single person who works as an artist at Disney/Pixar or their competitors is technically ripping off the SMALL handful of artists who created those signature "house styles", but it's morally okay because uhhh... reasons, I guess. Because, again, they're being paid for it? OR, how about the situation with hit RPG Disco Elysium, where the original creators (including the artist!) now got ousted from their own company, which means that any sequels with visual fidelity to the original will be made by an artist expressedly hired to imitate the style of Alexander Rostov. Legally, one might add. Because IP law benefits independent artists... essentially never, and yet people keep calling for more copyright law to fix the problem of companies adding automated creators to their exploitation repertoire...

I abhor seeing AI art where I expect or want to see human art; it is more boring to me than most abstract art, and I've never bee a fan of abstract art. Imagine my annoyance when I keep having to defend it against people who hate it for the wrong reasons...

If you're mad at AI art because it's yet another example of automation being used to put people out of work, say that without inventing unrelated reasons to delegitimise it that are too complicated and unconvincing, like "it's lazy and/or derivative". If you (like me) are mad at AI art because being an artist is important for your ego and you don't like it when someone inputs a text prompt into it and now demands the same amount of respect and accolades that took you 20+ years to achieve through hard work and suffering, then say that, and don't weaken your argument with attacks on its artistic integrity that that inadvertently also throw entire swathes of previously uncontroversial art under the bus.

/rant

[Precure]

I think there is a big difference between imitating someone's art style and literally copying their art. With a reasoning such as this, we could have to accept even people who trace over other artists' works.

[Kish]

I'd just like to say that "previously uncontroversial" is...not a statement of virtue I find particularly compelling.

[Errorname]

I find it extremely odd that people with this position continuously imply that if these AIs were made WITH credit and/or compensation, then it would be a-OK.

To be clear, I very much don't, I think this tech is an anti-printing press and it's only function is to output worthless noise and it's making the internet tangibly so much worse to use.

The point of stressing that it is without credit, consent or compensation is because it means the bad thing is also powered by theft, and while I am not optimistic about being able to unring the larger existence of the tech, the most egregiously unethical usage patterns might be preventable

[Anymage]

I think there is a big difference between imitating someone's art style and literally copying their art. With a reasoning such as this, we could have to accept even people who trace over other artists' works.

The catalog of training data can be tossed out after the AI is trained, and the data isn't kept in memory in any way any of us could understand. Not any more than the data from reading this comic is encoded into the neural networks that are our brains. There are several ways that an artist could be clearly influenced by Rich yet still create a work transformative enough to be seen as a new thing. (Whether in the legal copyright sense, or the looser sense where fanfic might not fall afoul of copyright laws but still be seen by most people as not plagiarism.) AI isn't fundamentally that different.


The point of stressing that it is without credit, consent or compensation is because it means the bad thing is also powered by theft, and while I am not optimistic about being able to unring the larger existence of the tech, the most egregiously unethical usage patterns might be preventable

Some unethical actors have large amounts of money, and can legally buy large amounts of data to use for training. Others have no problem stealing and will happily scrape everywhere public discussion areas for training data all on their own. I'm curious what sorts of egregiously unethical actors you think will see restrictions as anything more than a speed bump.

[pyrefiend]

I find it extremely odd that people with this position continuously imply that if these AIs were made WITH credit and/or compensation, then it would be a-OK. For one thing, if the same technology had been made with the up-front cost of, say, buying up a large database of assets or paying one dollar per artwork to every person whose art is used - costs that are substantial but not exactly insurmountable for a corporation worth billions making a product that is expected to be very profitable - then the way AI art ostensibly threatens artist jobs and human creativity in general would still be a factor, and that's the thing people are actually mad about. For another thing, the creative industry is rife with examples of corporations interacting with individuals in a way that is technically legal and above-board, but still exploitative.

I agree with this so strongly. I can understand why people focus on the legal element of it, because it's simple and actionable. If MidJourney broke the law, then it should be possible to prove that, and there's ostensibly something that artists can do about it. But that's only a small part of how and why this technology is going to screw people over.

[137beth]

I am soooo excited for GPT-5 trained exclusively on Microsoft Word documents, Outlook emails, Microsoft Teams videos, and video games, all without paying the writers or artists a single penny. I'm even more excited for the endless parade of puff-pieces about how moral and ethical GPT-5 is for "respecting copyright" and "not stealing."

EDIT: And I'm especially looking forward to the authors of these puff pieces not knowing (or pretending not to know) the difference between copyright infringement and plagiarism.


I think there is a big difference between imitating someone's art style and literally copying their art. With a reasoning such as this, we could have to accept even people who trace over other artists' works.
Midjourney contains none of it's supposed "training data," making it very difficult to argue that it is "copying" anything. (Unlike GPT-4, which actually does contain copies of some of its training data, but hey, it's Microsoft so the rules don't apply to them.)

[Flyfly]

I'm afraid people who don't go "generative art" sometimes heavily misunderstand just how much genie is out of the bottle.

See, if some law is passed, and it makes Midjourney illegal, then that is that! Even if it happened say only in USA, it would cripple the company too much to continue. Servers are turned off, bye-bye, Midjourney! Should be simple, right?

But thing is, there is also Stable Diffusion. And Stable Diffusion can run locally. No need for a server, no need for an internet connection, you just download the whole thing! It doesn't even take up that much disk space. And it's been like this for multiple years now.

And that's the problem. There really is no putting this genie back - even if Stable Diffusion goes off, even if Goole Collabs go off, you can't just delete it from the many, many users' PCs who stored the whole thing already and can share it.

That's, of course, not to mention the whole "many countries have their own laws" thing. Chances are even if this were to work it would just mean that big corporation would outsource art to somewhere else.

People sometimes cite tech to "catch" AI generated art, but it doesn't actually work, at least the one's we've seen so far. They produce incredible amount of false positives. Plus, you would be able to avoid these with relative ease by just editing a given generated pic (which anyone actually using generative art in production would do anyway, even if only to remove 6th fingers). Even in a simple way: you can, say, upscale the generated pic, then downscale it back (which you should do anyway, even with normal art you want to work with higher resolution than your target resolution).

No matter how you slice, and no matter how to we feel of it, we are to live in this reality now.

And then there is also the other aspect people already mentioned here: being pro "ethical databases" is not the same as as being "anti-AI". It just means being pro-big-company (Google, Adobe, large photo banks that legally own a ton of pictures, etc). This is a coherent position to have, but the winning endgame here, for practical purposes, is that AI art is still an expected part of commercial art, but you have to pay Adobe subscription fee to do it. And I'll be honest, fighting to make sure artisits in the future Definitely have to pay Adobe is not a cause I am hyped about.

[Precure]

I'm pro-big-company then.

[KorvinStarmast]

All of these image generation AIs came about because computer scientists were originally building image classification engines. After training, you give as input a never-before-seen image and the engine comes up with a description for it, e.g., "An Impressionist-style painting of water lilies in a reflecting pond, reminiscent of a 19th Century French garden". Someone had the bright idea of running the engine in reverse -- swapping upstream and downstream -- and lo and behold when they typed in a description they got an image out that wasn't half bad.

After a few thousand training images -- much less five billion -- it's pretty much impossible to say where, if any place, a given image from the source data (or the words used to describe it) "lives". For an extra layer of complexity, sometimes they use the network to generate a very large set of data, erase the weights, and then use the generated data to create a brand new weight set. And computer scientists seem to be coming up with a bit of protection for artists. From The Hustle...

Nightshade, created by computer scientists at the University of Chicago, is a new way for artists to protect their work from unlicensed AI scraping. The tech tricks AI with subtle pixel changes, “poisoning” the data that the models interpret.

[Flyfly]

And computer scientists seem to be coming up with a bit of protection for artists. From The Hustle...

I mean... this doesn't really work as well as one may imagine it does. Firstly and obviously, it's a bit "too little too late".

But more importantly, this sort of "data poisoning" can be avoided pretty easily. I don't know how this one works exactly, but based on previous attempts at creating these 'protections' like Glaze:

Since they affect pixel data, it's actually surprisingly easy to break all the seals by doing image compression and resizing. Say, resizing picture to be bigger and smaller would just wash out all the small pixel alterations. In fact, since training models requires specific sizes and proportions (512x512,1024x1024), this resizing is already done by default. And they have to be small because...
...they affect pixel data! High values of protections just obviously affect how the picture looks (it's kinda ugly). I don't imagine a lot of artists want to smear what effectively is an uglifying filter over their work.

Point being, this ain't no solution.