Make your own AI-generated art

[Azaran]

That Seinfeld bit is pretty coherent.

Anyway, no one has really posted good art here because they're either still using a weak system or aren't using a good one well. If you want to see the state of the art, check out the Midjourney and Stable Diffusion Discord pages.

https://discord.com/invite/stablediffusion
https://discord.com/invite/midjourney

I don't know how people are coming up with photorealistic images on Midjourney. I've been trying out the new version with all kinds of extra prompts and my stuff still looks mediocre

[Azaran]

I can't get over this ChatGPT.

Prompt I made up: The Karmolians are a surrealist society of hyperdimensional beings. Write a script for their initiation ritual

Result:

The Karmolians are a surrealist society of hyperdimensional beings who value creativity and individuality above all else. As such, their initiation rituals are designed to challenge and inspire new members, helping them to unlock their full potential and become true members of the Karmolian community.

The initiation ritual begins at dawn, when the initiates are gathered together in a sacred chamber deep within the heart of the Karmolian city. The chamber is filled with strange, otherworldly artifacts and sculptures, which seem to shift and change form as the initiates watch.

As the initiates stand in awe, the leader of the Karmolians enters the chamber, carrying a glowing crystal in her hand. She approaches each initiate in turn, placing the crystal on their forehead and speaking a sacred incantation.

As she speaks, the initiates begin to feel a strange energy coursing through their bodies, as if the crystal is unlocking hidden powers within them. They feel their senses heighten, and their perception of reality shifts, allowing them to see into other dimensions and experience the world in new and fascinating ways.

With the incantation complete, the leader declares the initiates to be true members of the Karmolian society, and they are welcomed with cheers and applause from the other members. The initiates are then led on a tour of the city, where they are shown the incredible feats of creativity and artistry that the Karmolians are capable of.

Throughout their journey, the initiates are encouraged to challenge themselves and push the boundaries of their own creativity. They are given the freedom to explore and experiment, and to discover their own unique talents and abilities.

As they complete the initiation ritual, the initiates are filled with a sense of excitement and possibility, ready to embark on their journey as true members of the Karmolian society.

[Azaran]

Something to note too is ChatGPT for some reason initially refuses to give the requested info half the time, saying it can't browse the internet or has no info on the topics. But then if you try again it spits it out

[Sulphur]

Speaking of ChatGPT, this is fairly impressive stuff - you can get it to simulate a Linux shell interface.

What this means is even more impressive than the admittedly great result, because it's taken the training data for not just English language text, but also codified machine interfaces and rendered them usable within the same use case.

[Aja]

Aesthetics are important, but the most important part about art is the connection it creates between the artist and the viewer. Art is about communication. AI can get the aesthetics right, but until it's sentient, any art it creates is otherwise meaningless.

[Sulphur]

Quite untrue. Human beings generate meaning on demand regardless of who or what created something; at this point, GPT-3 can even give you a semi-plausible explanation for a piece of generated art. As long as we get sensationalist pieces like this, we can always prove ourselves to at least attach surface level meaning to most things.

[Cipheron]

As for generating novelty, one way that works really well is say you input 1000x1000 pixel images, and tell it to output the *same* 1000x1000 image. You wouldn't *think* that your training a machine to make novel images.

But those million inputs need to go through different layers of the neural network, and those can be a lot smaller than 1 million. Say the neural net goes through a layer with just 50 cells. Well, now that layer is only passing 50 values through, so the network has somehow learned to compress 1 million inputs down to 50, and we know that because it's able to re-create the original images from just those 50 inputs now. At that level of compression it's no longer sufficient to just approximate colors, it needs to learn deep structural facts about the data itself.

You can then just chop the neural network in half, and only give those 50 inputs as the inputs into the second half of the network. Then, those 50 values can become "sliders" that you can manipulate. This is how it works when you get those NN image generators that DO give you some levers, and facial features morph as you move the levers. No human decided which levers are going to exist, the neural network worked that out for us.

So you can then explore parts of this 50-dimension search space you created - areas where no human has searched, and no human would THINK to search. So while the original data was based on human or natural creativity, the generator allows you to search hypothetical areas that fit the "rules" the humans made, but no human thought process would ever think to look there.

[demagogue]

Lol.

[Sulphur]

As Roger Ebert once said, 'ceci n'est pas une pippee'.

[Azaran]

It can also reconstruct fragmentary texts. Give it some background on the text, then ask it to fill in the gaps, and it will come up with a plausible reconstruction

[lowenz]

Related

https://www.pcgamer.com/ai-art-gener...nd-very-wormy/

[Aja]

Quite untrue. Human beings generate meaning on demand regardless of who or what created something; at this point, GPT-3 can even give you a semi-plausible explanation for a piece of generated art. As long as we get sensationalist pieces like this, we can always prove ourselves to at least attach surface level meaning to most things.

The communication/connection aspect is still a really important part of art that can't by definition be part of art created by a nonsentient being. Once the machines develop intent, than all bets are off.

And no, I'm not jealous of AI art. I think it's neat and cool, but the answer to the question of whether it will replace human-generated art is: only if it can do everything that human-generated art can. And for now that's not the case.

[Cipheron]

The communication/connection aspect is still a really important part of art that can't by definition be part of art created by a nonsentient being. Once the machines develop intent, than all bets are off.

And no, I'm not jealous of AI art. I think it's neat and cool, but the answer to the question of whether it will replace human-generated art is: only if it can do everything that human-generated art can. And for now that's not the case.

I think the conveyance of ideas and feelings is the crux here, but i have a problem with the use of the term intent or artistic intent here. Long before the discussion on AI art, "intent" has been a code-word for dismissing vast swathes of popular art as not being "True Art". The "True Art" crowd aren't about excluding machine-created art, but most actual art made by real people. See for example the big argument about whether Bob Ross landscapes are "true art".

And "Intent" could be used as a goal-post-shifting / No-True-Scotsman type thing. "never play chess"=>"never beat a human"=>"never beat a grand master", "never write a sonnet"=>"never write a 'proper' sonnet" etc. Even if the AIs become able to perfectly embed ideas and feelings into a work, then we can still "No-True-Scotsman" them by just saying "yes but they lacked PROPER intent so it's still invalid", since 'intent' is also a synonym for consciousness itself. So "intent" sort of fails as point here, because in the context of AI art, it's a completely non-falsifiable claim.

Also, a machine intelligence may have intent that's completely incomprehensible to us. We're not really talking about that at all, we're talking about the machine knowing about humans, and us looking at a painting and going "oh that's conveying that human thing we do, how clever". So since they're not humans, they can't know "human stuff" automatically and thus be able to spit it back at us unless we tell them about it, and explain better what exactly we want. So that particular lack isn't really telling us about the limits of AI at all, as much as it says just how bad we are at explaining this stuff to the machine as well.

[Aja]

I guess I'm looking at this more from the perspective of: why do I create art? I'm no great artist, I have no acclaim, yet I've spent the last 15 years creating and performing music with a strange, insatiable desire to keep doing it. Why? Part of it is wanting recognition and praise. Part of it is a sense of satisfaction in having created something. Maybe part of it is wanting to leave a legacy. And part of it is the feeling that I need to bring into the world something that hasn't hitherto existed. I want people to hear what I've made, to interpret it in their own way, and to feel less alone. And these feelings absolutely influence how I experience other peoples' art.

So I don't think it's a fallacy to say that the most important things to me about art are things that nonsentient beings aren't capable of and that, therefore, nonsentient art can never be as meaningful to me in that way. I shouldn't have used the word "meaningless" because I don't mean to say that AI art can't be meaningful. But intent and authorship are important to me.

Here in Edmonton we have an elephant in the zoo who paints. The paintings are fine aesthetically, but what makes them moving to me is the possibility that this animal might be expressing its desires and personality through them. If it turns out it's not, then it's just boring. If we shift the conversation toward whether the AI is trying to express itself, then suddenly it becomes very interesting.

[demagogue]

My first thought on that last example is, especially for the region I focus my research on, Southeast Asia, elephant painting has a long history of being associated with exploitation of the animal. But on that note, Adorno famously wrote a book called "The Culture Industry" about how the production of mass popular cultural goods in modern societies is almost inevitably exploitative for the artists involved, and how difficult it is to have an authentic scene thrive driven only by the art itself. AI Art has just thrown its hat into the ring of what was already a pretty messed up system.

[Azaran]

Over the next few years countless new people will try their hand at publishing books. They'll write coherent and informative (though perhaps rehashed) books on topics they know little about, using ChatGPT.

Here is a tool that can write coherent and unique dialogues, theatre plays, ceremonies & rituals, poems, hymns, stories, book reviews, blog posts, essays, &c, in a matter of seconds. Should be interesting to see what happens.

I also imagine the days of students having to write essays or complete written homework have come to an end (as it's impossible to prove that the student used an AI). I wonder what schools will do now that this tech is out there for all?

[Tocky]

Ultimately it's up to each of us to decide what is or is not art. Since it is subjective there is no right answer. It's only what is right for us. A lot of people thought Picasso was art. I think any AI at all beats that kids refrigerator art style. That doesn't mean it beats most human art to me. It can still only mimic a style and not create one. But it does that well.

And I'm right. For me.

Also I was expecting more posting of the art it did produce. Color me disappointed. Seriously. This is what color me disappointed produced. And I am. Amused though.



P.S. Azaran, can you imagine the lack of imagination in the next generation when no effort is put into anything? The horror...

Just for contrast here is an actual painting by an actual person.

[demagogue]

Looking at that, just as a public service announcement, the first thing you have to know about AI art is there are certain things it routinely botches, including messed up fingers and arms, legs and feet, eyes, asymmetrical faces, watermarks, low resolution parts, etc. Then the next thing you have to know is that you fix these things with a routine mix of positive prompts ("symmetrical face", "4K resolution") and negative prompts ("too many fingers", "watermarks").

It's become so routine that there are model background prompts to eliminate artifacts that should just be included no matter what you're making. Once you get it right, you can basically get rid of most artifacts.

[zacharias]

Keep in mind that 99% of human made art is formulaic or downright copycat, comedy included.

No real argument from me here. No offence, but I’m not quite sure what your point is? I’d say 90-95% of everything is crap, not just art. An even smaller percentage is exceptional or original, yeah let’s say 1%. (I don’t have a problem with relatively conventional art that is well executed though. Competence is valuable/underrated.) AI art seems like it can do better than the 99%. But my point is can it do as well or better than the 1%? The majority won’t care, I do concede that point.

I think comedy would be a weak point for AI. I just can’t see it doing comedy well, bar absurdist stuff perhaps.

And dema, I assume your ‘lol’ post was directed at me? The ‘not real art’ stuff. C’mon man, those prompts are so loaded as to already making the point they want to by themselves. But you basically said to me ‘check out the mid journey discord if you don’t think these are good’, as if there was no possibility the mid journey art might not be good? N.B. I don’t have time to play with the software right now, but I looked at the discord and community showcase page. My takeaway: astoundingly polished and pretty images, looks like something you would see in spectrum best of contemporary fantasy art. What I don’t really see though, are images that have a decodable or intelligent concept upon which one can ruminate or cerebrally engage. I may be wrong; but I expect someone like you to do more than point me to the discord channel. Pick an image, any 1 image which has an intelligent concept which is decodable and with which I can engage, and point me to it? Genuinely curious.

Sulphur’s link was cool. Genuinely weird and creepy images. Not sure if I’d think about it much after the viewing though.

Someone mentioned the ‘not real art’ thing previously as a controlling thing. Yes, that kind of gatekeeper thing was very real in the art world for centuries now. I assume it was to control the market/economics. This feels almost the opposite, a free for all.

My worry is this stuff just feeds into our instant gratification impulses in all things. Human produced art will be an increasingly niche thing. And who will bother to acquire any real skills in image making anymore? Either conceptual or practical. And I think that is a real loss. I don’t find it particularly funny.

[demagogue]

And dema, I assume your ‘lol’ post was directed at me? The ‘not real art’ stuff.

The lol was a joke that fit in the context of what all of us here are posting about in this thread, yes also in the context of what you posted, but no it wasn't exactly directed at you or your post per se. It's just a quick joke in the context of where the thread and the scene generally are right now.

But to respond to what you said anyway... One, the first thing I think about AI art is that it's another glorified tool artists can use to realize their vision. 95% of it is soulless sludge I think because 95% of all art is soulless sludge--except now it's very polished and accomplished soulless sludge--because the people writing the prompts don't have very good concepts to begin with.

But there are people that know how to make good concepts, and their art is engaging, not because it's AI art per se, but because they know how to get the prompts to work with their concept to make something inspired and interesting, like any art tool. Right now I think the best work is coming from people who have the minds of artists but not the technical capacity to get their concepts realized otherwise, or maybe people that do have the technical capacity but they like this outlet as well.

As for showing some good examples... To put that in context, my AI Art folder easily has over 4000 pieces already. There's such a flood of it. I think it's like a new kind of form where a person would need a guide to appreciate it, like someone getting into jazz or black and white photography. But the difference here is... you know what pieces to show to people to get them into jazz because it's a century-old art form by now. But this stuff is so new, and it's in such rapid flux, that it hasn't even had the time to coalesce into what form it is as a form. The norms and rules are changing as we speak. I think one can't even give a good overview over what's going on, for someone that just wants the thumbnail version, until there's been enough time for things to shake out for a clear thumbnail version to even emerge.

[zacharias]

Ok well, thanks for the response. It kinda gives me an idea where you’re coming from at least.

[Aja]

Sometimes a thing is valuable because a human made it. Imagine you had two pieces of furniture, tables. One was created by a craftsperson, the other by a machine. When you look at the hand-cut dovetails and hand-scraped, glass-smooth surface, they may look identical to the machine's, and if no one told you which is which, you probably would see them as equal. But once you know which one is human-made, that one suddenly becomes the more valuable of the two.

Y'all are making me feel like I'm arguing on the wrong side of history, but I just don't see how these certain important aspects of artmaking could be made obsolete by technology. The more interesting aspect to me is, as dema says, how artists can use AI to create new forms of art.

[Hit Deity]

I can't get over this ChatGPT.

Here's a ChatGPT "review" of the new Portal RTX.. pretty funny!
https://steamcommunity.com/profiles/...ended/2012840/

[demagogue]

Y'all are making me feel like I'm arguing on the wrong side of history, but I just don't see how these certain important aspects of artmaking could be made obsolete by technology.

It's also interesting how much this debate sounds like the kinds of things Walter Benjamin was saying 90 years ago in The Work of Art in the Age of Mechanical Reproduction about the tension between technology and human authenticity. Plus ça change...

[SubvertizingOrg]

Shut the fuck up all of you.