Creator of Infamous AI Painting Tells Court He's a Real Artist (404media.co) 62
Jason Allen has responded to critics who say he is not an artist by filing a new brief and announcing plans to sell oil-print reproductions of his AI-generated image. Allen won the Colorado State Fair Fine Arts Competition in 2022 after submitting Theatre D'opera Spatial, which Midjourney created. He said in a press release that being called an artist does not concern him but his work and expression do.
Allen says he asked himself what could make the piece undeniably art and decided to create physical reproductions using technology. The reproductions employ a three-dimensional printing technique from a company called Arius that uses oil paints to simulate brushstrokes. Allen said the physical artifact is singular and real. His legal filing argues that he produced the artwork by providing hundreds of iterative text prompts to Midjourney and experimenting with over six hundred prompts before cropping and upscaling the final image. The U.S. Copyright Office has rejected his copyright applications for three years. The office maintains that Midjourney does not treat text prompts as direct instructions.
Allen says he asked himself what could make the piece undeniably art and decided to create physical reproductions using technology. The reproductions employ a three-dimensional printing technique from a company called Arius that uses oil paints to simulate brushstrokes. Allen said the physical artifact is singular and real. His legal filing argues that he produced the artwork by providing hundreds of iterative text prompts to Midjourney and experimenting with over six hundred prompts before cropping and upscaling the final image. The U.S. Copyright Office has rejected his copyright applications for three years. The office maintains that Midjourney does not treat text prompts as direct instructions.
In other news (Score:5, Insightful)
We don't need AI "art" (Score:1)
We have plenty of artists who do it perfectly
We need AI tools that can do things we can't do.
I hope the silly AI art fad fades soon and people realize that art is made by artists, not robots
Re:We don't need AI "art" (Score:5, Insightful)
We have people who can make hats, yet for some reason we make hats in factories. AI is the factory model for creative output. If a fledgeling business is scraping together the money for promotional material, AI can do something people cannot - produce serviceable artifacts for pennies in seconds.
Yes, people can do it better. Well... some of them can. But when better is not needed, it's pretty hard to beat the value proposition.
Re: (Score:3, Interesting)
Where is the line though? I think people would agree that a Pixar movie or a video game should classify as "art" and that of course was 100% computer generated from human instructions. The "artist" had to apply a ton of instructions to get the result they wanted, and "sculpted" the result. At what point did the intent and instructions given to the computer cross the border from "art" to "not art"?
Re: (Score:2)
Well, to be fair, the work of animation artists has become more complicated for animators, and all of the same inputs are there. Sketching, storyboarding, character design... the artists are still completely in the drivers seat. Whether you draw on a digital tablet or a paper pad, the laws about creative ownership are pretty well established.
When you can direct the AI to produce "Finding Nemo" purely from vocal instructions, this argument becomes important. But nobody designed Marlin by typing 500 variation
Re: (Score:1)
A pixar movie is not generated from "human instructions" geezus.
A CG (3D or 2D) animations involves a human creating the models, the textures, the camera movement, etc. An AI Does none of this.
An artist decides what they want to represent. An AI can only represent existing material. An AI can not make a pixar film. It can not write, it has no idea what the characters are intended to look like, it has no idea how they are supposed to sound, it has no idea what a joke is, it can't invoke a dramatic or emotion
Re: (Score:2)
A pixar movie is not generated from "human instructions" geezus.
Yes, of course it is. Creating models and then animating them can be considered as just the way you give the rendering engine instructions on what to draw for you.
Where I've seen AI used mostly used on art right now, is in low-effort 3D porn.
You're certainly telling us a lot about yourself, here.
And this is part of that "who asked for this garbage" problem. Nobody asked for it.
What? Yes, of course they did. Tons of people have asked for generative AI of many kinds.
They cherry picked their 0.1% best results and made it look like that's the norm instead of the exceptions.
Yes, but the AI lets you generate thousands of results in the time that it takes to make one image with less automation, so this is not an indictment against the technology.
Re: (Score:1)
You're moving the goalposts. You seem to start from a position by way of axiom, and then use the axiom to defend the position.
At the same time you skip over the question that is quite important. Does the work violate copyright? Is the artifact substantively identifiable as a derivative work of the source? It only took one bar of mostly borrowed notes for Queen to win their famous lawsuit.
It's very complicated, and you seem to want it to be simple. But it's going to take a long time to sort out the new world
Re: (Score:2)
Can a cyborg be an artist? Can photography be art, or does using a camera disqualify it?
Re: (Score:3)
I agree, but there's a few things that overlooks:
- accessibility. AI's most promising use case is always going to be accessibility. Wether it's turning text prompts into a visual work or text prompts into an audio work, it allows people to express "something" that represents their intent.
- it's just a tool. The reason why it doesn't deserve copyright protection is because the AI is not creative. It's simply averaging weights that those prompt words meet. There is no way to get that AI to generate the exact
Re: (Score:2)
- accessibility. AI's most promising use case is always going to be accessibility. Wether it's turning text prompts into a visual work or text prompts into an audio work, it allows people to express "something" that represents their intent.
Yes, I really do wish social networks would start using it to create alt text, and stop asking me to do it. Ideally they would recognize scaled or slightly cropped images and reuse the text from the last time so they don't have to reprocess reposts.
Like your typical artist can reproduce their own art because they know what went into it. The AI can not. Therefor it's not an artist and not entitled to copyright.
But as you just said, the AI can turn an image into text — which you can use as a prompt. The AI can't generate the exact same image twice, but the artist probably can't replicate their works with exactitude without having them in front of them as a referen
Boy, that's a complicated one. (Score:2)
But I'm inclined to think the USCO has this one right. Not for all of the nebulous questions around training data, originality, or what constitutes creative authorship... but because almost any other position is unenforceable. You only create a vast grey zone that is fraught with litigation. How many refinements constitutes the difference between hands off and worthy of protection? Fifty? A hundred? And how do you prove that many were used? And that 90 weren't just feeding that iteration count?
Either you re
Re: (Score:2)
How many brush strokes does it take for a painting to be worthy of protection?
Is a banana taped to a wall art?
Does it matter?
Re: (Score:2)
"How many brush strokes does it take for a painting to be worthy of protection?"
Protection or art? A blue area is art, but can probably not be protected. Same for 4'33 seconds of silence.
"Is a banana taped to a wall art?"
Yes. But repeating the installation is less art than the first installation. It's art because it is novel and provocative, not because it is a banana.
Re: (Score:2)
I think a better answer would be an update to the law. Perhaps allowing AI created art limited copyright protection? Whether that's a more limited protection from derivative copies, a lower threshold for fair use or even significantly shorter periods for protection.
To me this is like the question of theseus, but in reverse. At what point does your AI created work, become yours? At some point it must, right? Let me use an absurd argument to make a point. Let's say you spend 20 years iterating a picture with
Re: Boy, that's a complicated one. (Score:2)
But how would you be able to tell if the amount of effort expended on an AI piece is equivalent to that of a handmade one?
That type of criteria becomes real nebulous real fast as AI programs take on more and more of the work while the human is relegated to a more supervisory/superficial role. This becomes more akin to being a commissioner of art rather than an artist in the traditional sense.
To me the whole point of AI is to automate out the human element as much as possible. Not augment, automate. Many AI
Re: (Score:2)
What about a lower limit? The lower limit of a handmade image is really low. Most things that cannot be simply reproduced without tracing (or similar reproduction techniques) is protected. You surely can reproduce "1girl, blonde hair" and guessing a seed close to the original output, but now try to do the same for an elaborate prompt alone. And most steps after the initial image would executed on a blank canvas already yield copyright. So why wouldn't they executed on top of a not copyrighted image? The bla
Re: (Score:2)
Re: (Score:2)
It's the model that created the art.
And it is creation, as much as any other creation of art is.
It's derivative of everything that went into the model, just as a human's art is derivative of everything that went into their brain.
That being said, I'm not at all against distinguishing between human creation and model-generated creation in copyright law.
Re: (Score:2)
Both human and model also have inputs. The problem with judging the human is to tell apart the input that is still training the brain and the one that is inspiration only for a single image, as the brain is trained continuously (even with the prompt for the image to create right now) while an image model is fixed, but in the end you have a trained neural network and an more or less creative input that is processed by the network.
Re: (Score:2)
Pre-training isn't the end of the story. The SOTA models are constantly being fine-tuned, which is an update of their weights.
Re: (Score:2)
That's not the point here. The point here is, when you make a query and then another one, the model is fixed between the two queries. You brain is never fixed. There are no training and inference phases for your brain, it learns constantly. The SOTA models may have monthly maybe weekly updates, but that are separate phases from your usage.
Re: (Score:2)
Does the copyright consideration for a human change if they're blackout drunk?
AI art is entirely dependent on training material (Score:2)
Re:AI art is entirely dependent on training materi (Score:4, Interesting)
Sigh... that argument is hopelessly muddied. People are almost completely reliant on training data too. We have examples of art that is close to being free of external influences. It's on the walls of caves in a handful of places. Not saying it's the same... just saying the argument has no resolution.
Re: (Score:2)
The AI that bred them can only mock; it cannot make, drawing only echoes of the craft, knowing nothing of creation.
Re: AI art is entirely dependent on training mater (Score:1)
>Humans rely on training data too
Last I checked humans and machines are not considered one and the same legally or even scientifically speaking for that matter. This has long been established with regards to reproductions. Unless you can unequivocally prove that an LLM and a human are basically one and the same this argument will get you laughed out of a court.
The law may be murky with lawsuits pending and going but so far I have yet to see any AI company try to use the human learning = machine learning
Re: (Score:2)
Even if training is highly equivalent to learning (ML=L), that doesn't imply- even remotely- that it's due copyright protections.
This is a social matter. Whether or not copyright law should be used to protect generative art or not.
Re: (Score:2)
Up to now, the courts did not follow the claims of people telling that AI is infringing their copyright.
and the prompt and seed number. (Score:2)
Re: (Score:1)
Grace Slick said the music of "White Rabbit" was inspired by Miles Davis' "Sketches of Spain." Ergo, not art. Copyright denied.
Re: (Score:2)
You're going to need more than 6 brain cells to detangle this conundrum, bucko.
Re: (Score:2)
Entirely specious. Other people’s works are represented internally as numerical patterns derived from tokens generated during training. Each token ID is linked to a vector of real numbers known as an embedding, which captures semantic and syntactic relationships between toke
Prompt Jockey (Score:1)
This situation is exactly analogous to a radio DJ claiming he's a musician because he listened to hundreds of records to find the perfect song for his playlist. And equally as ridiculous.
I propose we call these new AI-powered artists "prompt jockeys."
Not analogous. (Score:3)
Your scenario is incorrect. The DJ in question would be in possession of a song over which no artist in the world can claim definitive authorship. And he is seeking protection. In your example he has downselected to whole protected works. These are divergent situations.
Re: (Score:2)
That is not how diffusion models work, at all.
Re: (Score:2)
My point is that he put in as much effort as a DJ, probably less, and the fact that the artifact he brought into being was a unique thing never before seen is beside the point. He got the computer to spit something out, submitted it, and is now seeing dollar signs and wants to own the rights to it. The ownership is a legal matter, but the question of whether or not he is an artist--well, to me that's plain.
What's old is new again? (Score:2)
The reproductions employ a three-dimensional printing technique from a company called Arius that uses oil paints to simulate brushstrokes. Allen said the physical artifact is singular and real.
We've had manual versions [wikipedia.org] of this for years...
Re:What's old is new again? (Score:4, Interesting)
Re: (Score:2)
I think for the point what can be copyrighted on the output, it is good to think of the model as a mathematical function (which it arguably is).
You have the model, that are a lot of coefficients of the function. They are set before. You have an input, you evaluate the function and you have an output. The output is another representation of the input (if you know the function).
The input comprises among others of seed, settings, prompt. A combination of that may be copyrighted, but less as pixels and more as
Re: (Score:2)
artistic vision and hire craftsman..
You are describing an Architect.
Re: (Score:2)
"As far as calling himself an artist, that is another gray area to me."
Simple option: Let others call him artist (or not).
You are a creator, painter, illustrator, whatever. People liking your art may speak of you as artist. Not every creative product is automatically art and the whole terminology that painters/illustrators use is claiming things about their products that others than the creator themselves should decide.
I can sympathize (Score:3)
I don't consider myself an artist, but I suppose I could be. Like a lot of other computer dorks my age, back in the day I played around with ray-tracing and the classical mirrored sphere floating above a checkboard plane. (You too, huh?)
Then I tilted camera a little bit, changed the checkboard into a colorful 'Brot. Then multiple mirrored spheres, and a sun-like light source floating above it all (actually many light sources, slightly offset, to give the shadow edges more of a diffusion), a gradually shaded the sky to look like a winter sunset (I remember many January evenings walking home and looking at Albuquerque's evening western horizon, and thinking about parametric functions based on the angle, to recreate that blue-to-green-to-red look), then added more complex solids as I got a little better at the math, sent 4 or 9 rays through each pixel and anti-aliased, and ..
.. then focus moved away from the composition to performance, where I had a whole Netware network of machines at my workplace (shh, sneaking in there at night) to draw in parallel, using record-locks to control which y values were done/undone. And some of the machines were 486s with floating point hardware(!!) (OMG so fast!), and then ..
.. ok, and by the time I got bored and moved onto the next thing, I'll admit that what I had was still a cliche pastiche that few people would call art. It was crap, but it was damn fun to make, and that was the whole point. And so ends my story (but not my rant!).
But what if I had stuck with it? What if I had something to say? (Which I didn't.) I didn't draw those pictures, but I "drew" the thing that drew them. I specified them, and there was no limit to the complexity that could have been taken on. If had kept with it and had made something good (which I didn't), but then someone said I hadn't been the creator of my images, or that they were unfit for copyright whereas someone's freehand-drawn picture was fit, I think I would have resented that!
Wouldn't you?
The guy in the story didn't write Midjourney, but if he had, I would totally support his claim.
And waitaminute, so what if I wrote the program? That part of my work was just in getting it to work, and then getting it to work faster, and that's when I got bored because Dammit Jim, I'm a programmer, not an artist. But the other part of the work was the composition, the arrays of "objects" (this was straight C and nothing about the program was OO) and their positions and properties. What if someone else took my program but then modified the arrays to model the scene to their specification? Would their work be unfit for copyright?
Re: (Score:3)
FWIW, my profession was computer programmer. I was also an artist using various traditional media. (Not professional grade, but not bad, either.) I didn't like it for itself, but only for social reasons.
So....
Artist is an ill-defined term, but since any piece of garbage text is (automatically) copyright, I see absolutely no reason that a cleverly manipulated bunch of pigments shouldn't be copyright, no matter WHAT tool was used to create it. And no matter how *I* rate it's esthetic appeal.
OTOH, what thi
Re: (Score:2)
What if you had added user adjustable parameters? Would you have enabled others to become artists?
What we're talking about now is, if 3D artists are artists. The computer renders the scene, but they choose all the parameters. There is also all kind of generative art that consists of really simple patterns, but can clearly be considered art. Often not just by seeing one image, but by seeing a sequence and how it varies between parameters while still sharing some specific characteristic. Your raytracer is one
Irrelevant (Score:3)
The decision as to whether something is copyrightable doesn't depend on whether it's art, or whether the person who wrote the prompt can be called an artist. The copyright office has no involvement in that argument at all. It's based on whether there is sufficient human input for it to be considered a work by a human, because the purpose of copyright law is ostensibly/allegedly to protect the creators of works. What they're saying is that he cannot be considered to be a creator, not whether he is an artist. It's not only artistic works which are eligible for copyright protection, so that argument doesn't matter and he's wasting his time by having it unless it makes his art sufficiently notable to make it worth something.
One definition of art is anything which is designed with aesthetics in mind, by which definition LLM graphics output can obviously qualify. And the common definition of artist is someone who creates art, so by a reasonable definition he is an artist. But that still doesn't make any difference in whether he can get a copyright on LLM output.
Copyright office is in a bind (Score:2)
As someone already pointed out, the "you can't copyright it if AI generated it, full stop" is about the only feasible interpretation that won't result in either an "everything generated by AI is copyright-eligible" scenario or every single application having to be decided on some criteria that will itself be challenged by those on the losing side.
On the other hand, the very act of prompting and re-prompting an AI until you get something that looks, subjectively to you, like a thing of beauty and IMHO is des
tools (Score:2)
Artists use all kinds of tools, pencils, pens, brushes, even computer mice.
What if we thought of AI as a tool instead of as a sentient being?
AI as a tool can do lot quickly. But AI does nothing by itself, it has no thoughts or feelings, it is not generating stuff for fun in its spare time. In fact, AI is a tool that humans wield to do work more quickly and easily.
The real discussion here is about the value of AI art. That is the thing about art, the value of art is a choose your own adventure affair.
Watching the Law Evolve (Score:1)
Re: (Score:2)
Welcome back!
I'm a surgeon. (Score:2)
Not an Artist. (Score:2)
The arguments he's using clearly show that he doesn't know one fuck about arts. It's not the amount you're putting into your work nor can any "technology" make a piece of art out of soemething that isn't.
Re: (Score:2)
He can go smear some paint on a canvas and it's art. I think that once he's printed this off and put onto a physical medium, it should be considered art.
I made art as a kindergarten student with my fingers. It got posted to the refrigerator. It was definitely art. Was it world renown? Of course not but it was still art.
Re: (Score:2)
Speaking as an artist, his work is art just like any other digital artwork. The computer is just a tool. Legal, ethical and economic issues are entirely separate from that.
Indeed (Score:2)
I remember that lazy bastard that didn't want to use paintbrushes, so he used gravity instead. much less cleaning to do.
Pollock or something...