Stablediffusion, Midjourney and Deviantart face class action copyright lawsuit over image generation AI

[Formless]

Somehow, it wasn't until Corridor Crew's lawyer put up a Legaleagle style video that I learned about this one. But for those who prefer text:

Three artists have filed a copyright lawsuit against the creators of Stable Diffusion, Midjourney, and DreamUp, DeviantArt’s AI image generator. Sarah Andersen, Kelly McKernan, Karla Ortiz, and their attorney claim that these programs have infringed the copyright of “millions of artists” by training their algorithm on their work without permission.

The Midjourney founder recently admitted to using “hundreds of millions of images” without their authors’ consent to train the image generator’s AI. And now, his company and the two others could face legal consequences.

Lawyer Matthew Butterick is an artist himself, and he teamed up with litigators Brian Clark and Laura Matson of Lockridge Grindal Nauen P.L.L.P. On behalf of Andersen, McKernan, and Ortiz, he filed the lawsuit against Stability AI (the company behind Stable Diffusion), DeviantArt, and Midjourney.
Karla Ortiz
@kortizart
1/ As I learned more about how the deeply exploitative AI media models practices I realized there was no legal precedent to set this right. Let’s change that.
Read more about our class action lawsuit, including how to contact the firm here:
https://t.co/yvX4YZMfrG
8:14 PM · Jan 14, 2023
4.2K
1.2K

“Stable Diffusion contains unauthorized copies of millions—and possibly billions—of copyrighted images”, Butterick writes in a blog post. “These copies were made without the knowledge or consent of the artists.” He argues that, even if we assume the nominal damages of only $1 per image, this misappropriation would add up to around $5 billion!

In his explanation, Butterick wrongfully notes that Stable Diffusion is a “21st-cen­tury col­lage tool” and his lawsuit has been criticized over this. Stable Diffusion Frivolous writes that the diffusion model doesn’t store images, so it doesn’t collage them. It rather learns data distributions and then generates new work from them.

Butterick notes that the resulting images (“collages”) from AI image generators “may or may not outwardly resemble the training images.” While they’re not exactly collages, he might still be onto something here. Although small, there still is a possibility of AI-generated art resembling previously made artwork. With the right prompts and artistic style applied, you might actually generate an image that’s too similar to a work of an artist. It reminds me a bit of the “Infinite monkey theorem,” but it’s not impossible. And what do we do then? Does the original artist hold the copyright, or is it the person who created it using a text-to-image generator?

Scraping millions of images without artists’ consent is, in my opinion, a wrong approach, at least on the moral side. Many artists argue against using their work to train algorithms, but there are also those with arguments that support it. Here’s an interesting thought from an earlier article about AI training:

“If an art student studies all of Picasso’s 10,000 paintings and then creates a new painting that is clearly based on them, we call this the advancement of culture. The same is true if a writer uses a word that was coined by Shakespeare, or if a graffitist is clearly inspired by Shepard Fairey.”

In his article Patterns, Culture, and Theft, Seth Goddin argues that “taking an idea isn’t theft; [it] is an oxymoron.” He argues that this is how culture evolves and that “ideas belong to all of us.” So, we can’t say whether the judges will rely on arguments like this one or the ones stated by the artists who feel robbed of their work and years of learning and perfecting. But it’s definitely a topic to think about and to make it more defined in legal terms as soon as possible.

Did you know that massive data scraping is standard practice in the AI industry? And that the way they traditionally get away with it is to have a non-profit do the scraping for "research" purposes, which they then claim is fair use, then resell the data to commercial entities as happened here? Its actually been termed "data laundering", and I'm surprised this practice hasn't been shut down as an abuse of Fair Use doctrine already when its known those research organizations are nothing but commercial fronts for shit like this. The practice extends to all kinds of data, not just images used in image generation AI. Also, something the lawyer mentioned that the article did not, the way the AI works it doesn't exactly store copies of the images in the traditional sense, but what it does store is close enough that its kind of like how you can't claim you aren't storing a copy because its in a compressed format. For legal purposes, the plaintiffs will argue that the neural network is still intended to infringe on the copyrights of the artists by design.

As a user of Deviantart, I say its about goddamn time. The first time the community learned that the website trained an AI on all of our art was when they announced the AI was ready to use. The put a fig leaf on it just after the users revolted en mass, but all they did was let you put a tag on your images telling the website and external sites not to include this image in future data scraping. As far as I know, all of my own art was probably used to train DreamUp. Suffice to say, tons of artists left the platform over it, as its not like Deviantart's owners are well liked by the community to begin with. But that's a long story for another time.

[bilateralrope]

This is going to get interesting. First the lawsuits. Then politicians getting involved if they don't like the outcome of the lawsuits, with all the lobbying that comes whenever a law will have a significant impact on large businesses. Then probably some jurisdictions being AI-friendly while others are artist-friendly.

[Ralin]

So...doesn't every human artist maintain a mental catalogue of artwork they've seen and studied that they draw on when they create something? Isn't that a big part of how artists in general learn?

[Formless]

No, the difference is that human artists learn by process, not by copying. You need to get that muscle memory for your tools down before you can even try copying someone else's art style, not to mention the ability to envision the way the art will look like on a canvase, which involves learning how to transform a 3d view of the world that humans perceive into a 2d view of it. Perspective is not something you can learn by copying other people. A lot of people defend AI artwork by using that argument, but it is a misunderstanding or even a strawman. I can study a Picasso all day and never understand how to paint a Picasso. I would need to actually study the artistic techniques of Picasso before I could begin making art that looks like his. That's why an artist talking about and analyzing artwork to critique, study, or take inspiration from talk about it in a different way than the rest of us: they already have the skill to see beyond the exact image and instead focus on specific compositional techniques to emulate, whether its some trick of the lighting, texture, or form. No one wants to be called a plagiarist. Plus a human artist has a lifetime of experience with the real world to draw from whenever they draw or paint an image. The AI only has other people's work to draw from. And these AI don't know how to pick up a pencil or paintbrush and make art from first principles or even understand geometry. Their learning solely comprises a complicated algorithm for turning images they have already seen into noise and noise back into as close of an image from their training data as that as they can based on a text parser.

It is a completely inhuman way of generating an image.

[bilateralrope]

So...doesn't every human artist maintain a mental catalogue of artwork they've seen and studied that they draw on when they create something? Isn't that a big part of how artists in general learn?

They do. But when it's too similar to something someone else produced before, that invites a copyright lawsuit because it's too close to a copy of someone else's work.

Then there are the issues around why copyright laws exist. They exist to allow the artists (or whatever company the artist works for) to control who gets to make money off the artists work. To allow the artist/company to earn an income from their creative efforts, despite how easily anyone can copy it. That is what these AIs threaten.

I don't know what the answer should be. Only that there is potential for large companies on both sides of this fight once they start thinking about how the precedents set in this lawsuit over images will apply once AIs are capable of producing movies.

[Ralin]

It is a completely inhuman way of generating an image.

I know I shouldn't skip over all of your other arguments, but this stands out to me because of fucking course it is. It's inhuman because the image isn't being generated by a human. Obviously the AI will come at it through a very different process. That's sort of the point. Treating that as grounds for saying that AI generated art is inherently different to the point where feeding the AI examples of other art is exploitative unlike when human artists browse the freely available public website said art is posted on seems, uh. Questionable.

[Formless]

Let me ask you this: why defend something on the grounds that "humans do that too, don't they?" when you are just going to turn around and say "well, of course it isn't human, its a robot, so any difference between how it does things and how humans do things is irrelevant!"

That's dishonest.

[Ralin]

I think it's fair to compare the two at the points where they intersect, at least. Human artists spend lots of time viewing and analyzing art. A human artist browsing a bunch of (publicly viewable) art on DA and an art AI being fed a bunch of that same art seem at least partially comparable. The AI is 'just' (I realize that's a big just) processing a lot more of it.

[Shroom Man 777]

There's a big difference between how a human "processes" the art they study and the way an AI "processes" the art they sample.

There are cases of AI generated art included bits and pieces of the watermarks or signatures of the works they sampled lol.

Another example: A human reading up works and writing an essay inspired by the stuff they've read, with their own intent behind the words they're penning, is different from a machine process that samples other works (be they words or images) and jumbles them together.

The human mind's way of processing "source material" is different from the way AI samples it.

Conversely if a human writes an essay or makes a visual art that's too similar to something they've studied beforehand, there's legitimate grounds to sue them for plagiarism, right?

And anyway, there's also the whole "using others' art or writings for profit without their permission." Whether or not an AI does it, if a piece of your text or visual creation or music gets included in someone else's work and they profit from it without your permission, then?