Dreamy AI-generated geometric scenes mesmerize social media users

"This was the point where AI-generated art passed the Turing Test for me."

See full article...

 

[Hezio]

People have made art like this before. A good trick is to have two or more copies of David Wiesner's Free Fall, the whole book is one continuous geometrical illusion like these but with constant transitions. Here's some quick snapshots of my copies to illustrate:

View attachment 63290
View attachment 63291
View attachment 63292

(Pardon my messy desk, I swear I'm in the middle of cleaning it up.)

That was a really clever way to sell more books too.

Also messy desks imply used desks, which are better than unused clean desks.

 

[lil-irv]

Randomness is not creativity.

No it certainly isn't. But that's completely out of context with what I wrote.

What humans create is random in the sense there's a spontaneity and uniqueness about their creativity.

These AI results are very impressive but there's no actual creativity within AI. All the creativity is with the original artists, the programmers, and whatever resourceful ways it's used by humans.
We just can't ascribe these kinds of emotions and human traits to this generation of AI.

 

[Aurich]

I use Diffusion Bee on the M1 Mac for the user-friendly UI and specific optimisations. The latest stable version 1.7.4, or the beta one 2.x both support CobtrolNet natively now.

Ah, so it does. I didn't even look at any tutorials, just tried what felt obvious, it's pretty simple to use.

Our logo isn't the best input for subtle tricks, but there's 5 minutes of work. Eh, not exactly worth showing off.

I've always been of the opinion though that AI image tools are just tools. And you can wank with them or do interesting stuff with them. Back in the day the equivalent was Kai's Power Tools, and whatever that one you could generate landscapes with, spacing on the name.

They were fun, and often pretty boring, and sometimes someone would find an interesting or novel use.

This to me is novel, and a little interesting? It honestly loses the interest factor when you see how easy they are to generate. The lack of a human really planning it all out takes away some of the magic.

Edit: Should have included the prompt

A busy circuit board with glowing LED lights, (masterpiece:1.4), (best quality), (detailed)

 

[cbreak]

Randomness is not creativity.

Indeed. And Generative Models do have randomness, from the outside.

A model does not create anything at all, it is not creative in the slightest. All it does is convert random noise (which is generated from a seed, typically), into an image, guided by a prompt. The output is reproducible because of this deterministic nature, but changing the seed changes the whole image.

(And using a non-deterministic sampler will also change it)

 

[8x10 glosssy]

Looks like something M.C. Escher would have done.

 

[Ildatch]

Ya, but I never saw any that look like this.

You should expose yourself to more art. I recommend museums and art history books.

 

[barktrees]

You should expose yourself to more art. I recommend museums and art history books.

That's good advice for anybody. The book Aurich posted was nice.

 

[Hateful Species]

People have made art like this before. A good trick is to have two or more copies of David Wiesner's Free Fall, the whole book is one continuous geometrical illusion like these but with constant transitions. Here's some quick snapshots of my copies to illustrate:

View attachment 63290
View attachment 63291
View attachment 63292

(Pardon my messy desk, I swear I'm in the middle of cleaning it up.)

I am really impressed with the book/art layout. But the messy desk comment made me laugh so two points for that!

And what a great way to double the sales of your art book... 'You really need to buy two to get the full effect'.

Edit to add; What is that blue death ray gun looking thing?

 

[Aurich]

Edit to add; What is that blue death ray gun looking thing?

The short answer is it's a fancy flywheel-powered Nerf blaster that takes half length foam darts.

The more detailed answer is it's a Worker Nightingale, modded with a Black Raisins select fire kit to do semi, burst, or full auto, that will empty a 15 round mag in about 1.5 seconds, and fire darts around 130 feet per second.

It's not remotely dangerous, but you do want to wear eye protection if darts are flying around at that speed.

 

[Chuckstar]

Isn’t this sort of a fancier version of a photo mosaic? It’s just had the option of picking the two colors to be complementary items that it can stitch together at the borders.

 

[Ozy]

So, we know what you think of Pollock, but how do you feel about those no-talent sloppy impressionists?

Do you think they rolled dice to make their art? If you think impressionist art is 'random', then you don't understand art.

 

[Ozy]

No it certainly isn't. But that's completely out of context with what I wrote.

What humans create is random in the sense there's a spontaneity and uniqueness about their creativity.

These AI results are very impressive but there's no actual creativity within AI. All the creativity is with the original artists, the programmers, and whatever resourceful ways it's used by humans.
We just can't ascribe these kinds of emotions and human traits to this generation of AI.

Sure, AI is the tool to express, not the origin of the creativity. Is anyone claiming otherwise? It's merely that some people seem to think that using AI as a tool makes it impossible to express creativity, which seems like an arbitrary judgement.

 

[Edgar Allan Esquire]

It's been neat seeing the improved outputs with control net and the improved UI accessibility, but maybe it's because I've been rereading my horror section that I was expecting the image gallery to segue into Junji Ito's Uzumaki/Spiral territory where the body horror is a feature not a bug.

 

[Hydrargyrum]

The AI didn't instigate what makes these pieces original. That step came from a person choosing to use the software in a creative way with interesting inputs that hadn't been tried before. An machine-learning output being original isn't to the credit of the "AI"

That said, it seems to me that AI art generators have at least as many controllable settings, input choices, and artistic decisions to make as a camera with full manual controls. And it’s long settled that photography of objects that already exist in the world is deserving of copyright protection.

I don’t think denying the same protection to the output of an artistically controlled and prompted AI model is philosophically consistent with the position on photography.

Regardless, the model itself can’t claim the copyright IMO.

 

[lucubratory]

These are really cool. The "It looks bad, it's not art" argument was always doomed, eventually someone was going to make good art with the new medium and "Don't believe your lying eyes!" has never successfully persuaded anyone. A hundred years from now, people aren't going to look back and say "Jeez, we really dodged a bullet with the All AI Art Banned bill, I'm so glad those plucky copyright owners managed to strengthen IP rights to the extent that the technology never existed again. Now the world is a good place for artists, like it was before AI."

No, a hundred years from now there's going to be some dry academic article called "When Generative AI Wasn't Art" that informs a small number of art history nerds who even care that generative AI faced a fierce backlash when it was introduced as being soulless, non-creative, merely mechanical etc, with concerns about it destroying the livelihoods of existing artists. Just like what happened with photography.

 

[Hydrargyrum]

These are really cool. The "It looks bad, it's not art" argument was always doomed, eventually someone was going to make good art with the new medium and "Don't believe your lying eyes!" has never successfully persuaded anyone. A hundred years from now, people aren't going to look back and say "Jeez, we really dodged a bullet with the All AI Art Banned bill, I'm so glad those plucky copyright owners managed to strengthen IP rights to the extent that the technology never existed again. Now the world is a good place for artists, like it was before AI."

No, a hundred years from now there's going to be some dry academic article called "When Generative AI Wasn't Art" that informs a small number of art history nerds who even care that generative AI faced a fierce backlash when it was introduced as being soulless, non-creative, merely mechanical etc, with concerns about it destroying the livelihoods of existing artists. Just like what happened with photography.

And recorded music.

Recorded music has certainly reduced opportunities for jobbing musicians, while hugely increasing the payoff for the rare few hit acts… so there’s certainly going to be huge impact.

 

[flow783]

The problem is computers don't do random. No true random number generator.

AI can process and combine a painting with 100,000 other paintings and 100,000 other inputs and the output is only the result of all those inputs. Even an input like "but make it sad" is constrained to paintings labeled as such with the AI having no understanding of it.

A human is wonderfully imperfect. You can tell an artist to use 10 paintings with 10 constraints and they cannot limit themselves to that. They'll also have random thoughts and distractions and will hear and see things that will affect their output. The end result is random and not repeatable.

If you can use the same inputs and get the same output it certainly isn't creativity.

I don't know what it is but I certainly know what it isn't

None of your points hold water I'm afraid. First, the differential between 'truly random' and 'apparently random' is thin, the margin being such that mathematicians can't to this day cannot give you a strict definition of randomness, only collections of heuristics. A sequence of the same number 9 over and over may just be random; after all, in an unbounded, truly random sequence an arbitrarily long streak just may appear, however unlikely. But for practical purposes, our present pseudo-random number generators (PRNGs) by far exceed what any human could produce or discern in terms of unpredictability. Even math.rnd() is, thus, creative (in a very basal sense probably not fit for copyright legislation).

Second, I'd subscribe to your description of humans as "wonderfully imperfect" and the observation that a multiplicity of inner and outer inputs to the minds of men lead to irrepeatability of outputs, hence creative art. But then I can totally increase the number of inputs to my generative software; maybe I could feed it a stream of audio as heard in my place right now in the form of digits and letters or maybe words from a dictionary, thereby making the output dependent on my surroundings in the here and now, irrepeatable (and, hence, creative?).

Put another way, the sum of a number of numbers can be anything, and any number can be the sum of any number of numbers. Just because sums look simple and small integers look simple doesn't mean there isn't a highly complex world of integer sums out there mankind is still grappling with. One and two make three, yet three is more than the sum of of one and two.

 

[Chuckstar]

That said, it seems to me that AI art generators have at least as many controllable settings, input choices, and artistic decisions to make as a camera with full manual controls. And it’s long settled that photography of objects that already exist in the world is deserving of copyright protection.

I don’t think denying the same protection to the output of an artistically controlled and prompted AI model is philosophically consistent with the position on photography.

Regardless, the model itself can’t claim the copyright IMO.

The difficulty that will be run into is that requesting an artist “make a circuit board that looks like the Ars logo” and requesting an AI “make a circuit board that looks like the Ars logo” should result in the same level of copyright claim for the person doing the requesting. In the case of requesting that work from an artist, the artist gets the copyright, and the requester gets nothing. So why does the requester get a copyright when requesting the same thing from an AI?

 

[Aurich]

That said, it seems to me that AI art generators have at least as many controllable settings, input choices, and artistic decisions to make as a camera with full manual controls. And it’s long settled that photography of objects that already exist in the world is deserving of copyright protection.

I don't agree with this.

Unless you're reducing photography to the settings.

End of the day a human being has to pick up the camera, take it to the location, and capture a moment in time. There's an intentionality and human element that's utterly lacking from AI art.

I don't think that means AI art isn't interesting in its own ways, I just think trying to compare it to other things doesn't really serve all that helpful of a purpose.

 

[NYKevin]

The difficulty that will be run into is that requesting an artist “make a circuit board that looks like the Ars logo” and requesting an AI “make a circuit board that looks like the Ars logo” should result in the same level of copyright claim for the person doing the requesting. In the case of requesting that work from an artist, the artist gets the copyright, and the requester gets nothing. So why does the requester get a copyright when requesting the same thing from an AI?

Problem 1: It's more complicated than you make it sound. Depending on the specifics of the contract etc., it may be construed as a "work made for hire," in which case the copyright vests in the employer (or commissioner), and not the artist. Generally this only happens if you either have an employment contract or a contract that specifically uses the magic words "work made for hire," but it is definitely a thing in US copyright law (and the copyright laws of many other countries).

Problem 2: Consider the following procedure:

1. Go to Manhattan (or any similar location with lots of tall buildings) and take a photo of the cityscape (or, really, any interesting subject matter, this is just an example).
2. You own the copyright of that photo.
3. Put that photo into Stable Diffusion's img2img mode.
4. Set denoise to a value between 0.65 and 0.75 (or thereabouts - experiment as needed).
5. Input a prompt describing some fantastical or sci-fi variation of your image. Mention things like flying cars, hover bikes, etc., as appropriate.
6. The output will (probably) have significant copyrightable elements in common with your photo, so it is a derivative work of your photo, and nobody can use it without your permission. Regardless of whether the output can be independently copyrighted, you do have the de facto ability to control the output using copyright law.
7. But it will also have significant AI-generated elements, because of the relatively high denoise value. So you can get a copyright-controlled image that contains a significant amount of AI work.
8. You can now use inpainting to further refine specific portions of the image (e.g. to remove AI glitches and other artifacts), without destroying the copyrightable elements it has in common with your original photo.

 

[Fritzr]

That said, it seems to me that AI art generators have at least as many controllable settings, input choices, and artistic decisions to make as a camera with full manual controls. And it’s long settled that photography of objects that already exist in the world is deserving of copyright protection.

I don’t think denying the same protection to the output of an artistically controlled and prompted AI model is philosophically consistent with the position on photography.

Regardless, the model itself can’t claim the copyright IMO.

The controllability will improve. With more complex control of output, this will follow the route of photography and be recognized as an artist's tool and not a random image generator.

As for curation required, just ask any professional photographer "What percentage of photographs do you simply discard?". A large part of professional photography is knowing which photos to discard. Curation is a very important part of tool generated art.

 

[Fritzr]

The difficulty that will be run into is that requesting an artist “make a circuit board that looks like the Ars logo” and requesting an AI “make a circuit board that looks like the Ars logo” should result in the same level of copyright claim for the person doing the requesting. In the case of requesting that work from an artist, the artist gets the copyright, and the requester gets nothing. So why does the requester get a copyright when requesting the same thing from an AI?

You can ask a photographer to take a beautiful sunrise picture or you can click the shutter for yourself. In both cases the tool created the work, so why does the photo get copyright in both cases?

 

[Fritzr]

I don't agree with this.

Unless you're reducing photography to the settings.

End of the day a human being has to pick up the camera, take it to the location, and capture a moment in time. There's an intentionality and human element that's utterly lacking from AI art.

I don't think that means AI art isn't interesting in its own ways, I just think trying to compare it to other things doesn't really serve all that helpful of a purpose.

Are you saying an artist is not allowed to modify the AI's training in any way?

Jackson Pollock tossed paint into the exhaust stream of a jet engine and the jet exhaust painting is worth hundreds of thousands of dollars if not millions because a famous artist threw the paint and judged whether to keep or destroy the painting created by a jet engine.

 

[Fred Duck]

"This was the point where AI-generated art passed the Turing Test for me."

That just sounds dumb.

It is a matter of perspective. If instead of approaching the quote from a Kantian point of view, one instead adopts a stance closer to that of David Chalmers' Naturalistic dualism and examines the underlying socio-linguistic ramifications of the declaration, then it becomes apparent that Graham may be a little unclear on what the Turing Test is.

I, on the other hand, have acquired a very useful study guide from Libgen so I should be prepared when I take it.

 

[macosandlinux]

This stuff is getting really impressive. I know many of us have been shitting on AI art for a while but honestly, ask yourself, if you sent these back in time 10 years, would people look at them and go "wow what crap who made that mess?"

No, they wouldn't, not remotely. They'd be seen as reasonably impressive works of art. Perhaps nothing rare but you'd say, yeah that was made by a great digital painter!

Especially the spiral village, it's impressively Escher-esque.

It's basically what we use to 'mask' image areas in photographs, just no more the boring masked/unmasked, transparency/cloning. It's using the AI 'painter' function in a creative way. That's what's the impressive part, not what AI can do, but how users with ideas can make novel use of it.

 

[Zim the Vixen]

People have made art like this before. A good trick is to have two or more copies of David Wiesner's Free Fall, the whole book is one continuous geometrical illusion like these but with constant transitions. Here's some quick snapshots of my copies to illustrate:

View attachment 63290
View attachment 63291
View attachment 63292

(Pardon my messy desk, I swear I'm in the middle of cleaning it up.)

genuine question, what is that gun thing on your desk? looks like... a nerf gun but don't think it is? also what a delightful and beautiful book

 

[lil-irv]

Sure, AI is the tool to express, not the origin of the creativity. Is anyone claiming otherwise? It's merely that some people seem to think that using AI as a tool makes it impossible to express creativity, which seems like an arbitrary judgement.

I agree and I wasn't saying otherwise. My original comment wasn't well received so I didn't get my point across but it was that computers and AI are incapable of true creativity. *

AI is at a point where people don't trust it while some think it's some kind of magic.

I don't want to speak for you but it seems you're seeing the former and I'm seeing the latter. I think AI has a place not just in art, but can do a lot of good elsewhere. To my point: if people have the wrong ideas about AI they're likely to use it improperly and won't be able to judge its results properly.

(* not that randomness is creativity but the inability to be random necessarily means no spontaneity, no unpredictability, and none of the traits that allows for creativity)

 

[Aurich]

genuine question, what is that gun thing on your desk? looks like... a nerf gun but don't think it is? also what a delightful and beautiful book

It's a foam blaster, but it's not Nerf brand, it's made by a company called Worker (with third party upgrades I added), who got their start doing mods for Nerf blasters and then graduated to making their own original hardware.

Quoting myself from an earlier post where I answered this question:

The short answer is it's a fancy flywheel-powered Nerf blaster that takes half length foam darts.

The more detailed answer is it's a Worker Nightingale, modded with a Black Raisins select fire kit to do semi, burst, or full auto, that will empty a 15 round mag in about 1.5 seconds, and fire darts around 130 feet per second.

I have a bit of an obsession with high end foam flingers. Some people have asked me to do a write up for the front page, maybe I will at some point, but happy to answer questions about it, or start a Lounge thread or something too.

 

[85mm]

Why do people care if AI has creativity? AI is a tool directed by people. People have creativity, so works created by a person directing an AI can be creative.

 

[The Aloof Alot]

Have you ever seen any of M.C. Escher's work? It's very hallucinatory and geometric and this reminds me a bit of his work. (Not just the staircases, he did a lot more than staircases.) It also reminds me of Magritte's surreal stuff that plays with perspective and subjects.

I am not at all surprised that this set social media on fire. There's something very compelling about it.

Using mathematical patterns in art is almost as old as art itself. I think the excited reaction on social media tells more about social media addicts' lack of exposure to a large variety of human-generated art than AI suddenly "passing the Turing test".

None of the samples in the article look particularly original to me.

 

[Chuckstar]

Problem 1: It's more complicated than you make it sound. Depending on the specifics of the contract etc., it may be construed as a "work made for hire," in which case the copyright vests in the employer (or commissioner), and not the artist. Generally this only happens if you either have an employment contract or a contract that specifically uses the magic words "work made for hire," but it is definitely a thing in US copyright law (and the copyright laws of many other countries).

Problem 2: Consider the following procedure:

1. Go to Manhattan (or any similar location with lots of tall buildings) and take a photo of the cityscape (or, really, any interesting subject matter, this is just an example).
2. You own the copyright of that photo.
3. Put that photo into Stable Diffusion's img2img mode.
4. Set denoise to a value between 0.65 and 0.75 (or thereabouts - experiment as needed).
5. Input a prompt describing some fantastical or sci-fi variation of your image. Mention things like flying cars, hover bikes, etc., as appropriate.
6. The output will (probably) have significant copyrightable elements in common with your photo, so it is a derivative work of your photo, and nobody can use it without your permission. Regardless of whether the output can be independently copyrighted, you do have the de facto ability to control the output using copyright law.
7. But it will also have significant AI-generated elements, because of the relatively high denoise value. So you can get a copyright-controlled image that contains a significant amount of AI work.
8. You can now use inpainting to further refine specific portions of the image (e.g. to remove AI glitches and other artifacts), without destroying the copyrightable elements it has in common with your original photo.

Work for hire is irrelevant.

Work for hire is essentially an automatic transfer of copyright upon creation of the work. It’s not a different way of analyzing the work’s creation, only in who ends up with the intellectual property right.

In your example, you could do the same thing by handing your photo to a graphic artist vs an AI. The analysis of who did what and when is the same. The only difference is that the parts done by a non-human aren’t copyrightable, so in the case of the AI, intellectual property rights don’t attach to the parts not done by you that would have otherwise been done by the graphic artist. It’s still exactly the same thing I described above, you just added a bunch of steps to try to obfuscate it.

 

[Chuckstar]

You can ask a photographer to take a beautiful sunrise picture or you can click the shutter for yourself. In both cases the tool created the work, so why does the photo get copyright in both cases?

If you take a picture of a screen with an AI-generated image on it, you will also get the copyright of a photo taken of that particular angle of an image you didn’t create. Just like with a photo of a sunrise.

You don’t own the copyright on a sunrise. You only own the copyright on a particular angle, f-stop, etc. of a particular day’s sunrise.

I don’t think anyone disputes that if you take an AI-generated image and do stuff to it that is unique to your presentation of that image, that you own a copyright on those unique elements. What is in dispute is that it means you somehow own the copyright of the underlying AI-generated image. Your sunrise photo example provides no additional insight into that question, given what copyright does/doesn’t cover with regards to a photo of a sunrise.

 

[El Glob]

I've never seen a person make art like this before.

I envision a future in which someone called Giuseppe Arcimboldo, born in the 16th century, will devote his life to painting optical illusions. So magical will this future be that there will even be a whole category of this type of thing called "Trompe-l'œil art".

 

[Quisquis]

The difficulty that will be run into is that requesting an artist “make a circuit board that looks like the Ars logo” and requesting an AI “make a circuit board that looks like the Ars logo” should result in the same level of copyright claim for the person doing the requesting. In the case of requesting that work from an artist, the artist gets the copyright, and the requester gets nothing. So why does the requester get a copyright when requesting the same thing from an AI?

Because someone is putting in the majority of the creative input, and that input can rise to a level of creativity much higher than the minimum needed for a photo copyright, so I don't see why it should get special treatment.

 

[MrWalrus]

It's odd to me that an article about a (possible) work of art doesn't include the perspective of any artists or art experts. (And no, I definitely do not count either hypebro Paul Graham or 'a graphic designer named Trent' in either of those categories).

 

[Anton Longshot]

To all those who say generative AI cant do anything original. Derivative does not mean it can only regurgitate worse versions of the same content.

It does suck that its going to dilute and water down the overall content humans make, and that we likely wont be able to tell the difference between an M Night Shyamalan Film, and a generative piece of garbage, but im excited to see where this ride takes us.

Eye of the beholder / there's no accounting for taste.
Example: for me "generative piece of garbage" pretty well describes M Night Shyamalan's films.
I'm also excited to see where this ride takes us.

 

[lil-irv]

None of your points hold water I'm afraid. First, the differential between 'truly random' and 'apparently random' is thin, the margin being such that mathematicians can't to this day cannot give you a strict definition of randomness, only collections of heuristics. A sequence of the same number 9 over and over may just be random; after all, in an unbounded, truly random sequence an arbitrarily long streak just may appear, however unlikely. But for practical purposes, our present pseudo-random number generators (PRNGs) by far exceed what any human could produce or discern in terms of unpredictability. Even math.rnd() is, thus, creative (in a very basal sense probably not fit for copyright legislation).

Second, I'd subscribe to your description of humans as "wonderfully imperfect" and the observation that a multiplicity of inner and outer inputs to the minds of men lead to irrepeatability of outputs, hence creative art. But then I can totally increase the number of inputs to my generative software; maybe I could feed it a stream of audio as heard in my place right now in the form of digits and letters or maybe words from a dictionary, thereby making the output dependent on my surroundings in the here and now, irrepeatable (and, hence, creative?).

Put another way, the sum of a number of numbers can be anything, and any number can be the sum of any number of numbers. Just because sums look simple and small integers look simple doesn't mean there isn't a highly complex world of integer sums out there mankind is still grappling with. One and two make three, yet three is more than the sum of of one and two.

Some good points and a good thought experiment.
Strip it all to the basics.
Submit the same art to both and see what gets output.
Magically rewind with the only difference being the pseudo-random seed.
Repeat over and over.
Eventually the seed will reset and repeat. The computer's output will also start repeating.
What happens to the human?

 

[Chuckstar]

Because someone is putting in the majority of the creative input, and that input can rise to a level of creativity much higher than the minimum needed for a photo copyright, so I don't see why it should get special treatment.

What special treatment? I’m not seeing any special treatment.

A photo gives you copyright of the photograph, not of the underlying scene. Take a photo of a tree and someone else can also take a photo the same tree. You get a rights over your photo of the tree, not of the tree or all photos of the tree. Not even of necessarily of similar photos of the tree.

Seems like you’re the one who wants to grant special treatment of some type. You want your photo of the tree to grant you copyright of the tree, merely because no one else has copyright of the tree and hey, you put in a little effort and all.

 

[Quisquis]

What special treatment? I’m not seeing any special treatment.

A photo gives you copyright of the photograph, not of the underlying scene. Take a photo of a tree and someone else can also take a photo the same tree. You get a rights over your photo of the tree, not of the tree or all photos of the tree. Not even of necessarily of similar photos of the tree.

Seems like you’re the one who wants to grant special treatment of some type. You want your photo of the tree to grant you copyright of the tree.

That's a really dishonest interpretation...

The scene didn't exist before the prompt.