AI art generation

[Bohandas]

No, then you just need to pay $12/month (or only $100/year!) for the software license.

Oh right, I forgot that that "software as a service" scam existed. They probably would try to pull something like that instead of selling you a copy of the program honestly.

I am not enthused by this. I remember the world before social media turned a huge portion of human communication into something they could pump money out of. There was a lot to like about that. I'm not sure why I should be excited about another vast realm of human endeavor and enjoyment getting turned into a corporate fiefdom.

since when are music and publishing not corporate feifdoms? I was hoping that this technology would break those industries' power. The publishing industry may have taken a hit in the past decade or two but the music industry is more powerful than ever

[Bohandas]

I can't draw stickmen I like. AI art finally lets me do something artistic I enjoy, I've never had that feeling before in my life. I'm all for this.

Plus, even with the AI it can still take a while. Take a look at this:

https://steamuserimages-a.akamaihd.n...etterbox=false

it took me about an hour to successfully get the AI to turn this character (the mascot from a computer game I play that's set in the 1970's) into a tripping hippie without also making him unrecognizable in the process, and then it took me a further hour touching it up in GIMP to get it to the point where all his hair was the same color

[warty goblin]

Oh right, I forgot that that "software as a service" scam existed. They probably would try to pull something like that instead of selling you a copy of the program honestly.

I don't see any way they wouldn't do that, subscription services are where the money is at in software now, and at some point this is going to be a product.

since when are music and publishing not corporate feifdoms? I was hoping that this technology would break those industries' power. The publishing industry may have taken a hit in the past decade or two but the music industry is more powerful than ever

Sure, AI art might wreck the current music and publishing industries. All that means is we've replaced Penguin Random House and Virgin Records with Microsoft and Google and Tencent. At best that's a horizontal movement, and I'm incapable of getting excited about big tech companies owning another thing

But the other effect is going to be the massive loss in marginal value of any art form that gets a competitive AI. It takes humans weeks to years to write a book or a music album or a high quality painting, the AI is gonna generate all the music you could ever listen to for $10/month, and it'll do it in seconds. Wanna bet how many new bands are going to be able to launch careers against that? What company is going to hire an illustrator when ArtBot costs just $5000 a year for a commercial license*?

In a pure economic sense this is efficient, we get more art for less money, all while maximizing shareholder value. Its also going to deprive a lot of people the chance to do something creative that they love at an extremely high level. It'll probably keep a lot of people out of arts as a hobby as well; what's the point in learning to draw when ArtBot Basic does it better, immediately, and for free?

Seems a crying shame to me.




*ArtBot Inc retains sole copyright to all ArtBot output. A limited publication license is granted for ArtBot generated assets with subscription to ArtBot. Users must abide by the Terms of Service, which can change at any time. Upon termination of an ArtBot subscription, you lose all rights to ArtBot generated assets.

[Radar]

Sure, AI art might wreck the current music and publishing industries. All that means is we've replaced Penguin Random House and Virgin Records with Microsoft and Google and Tencent. At best that's a horizontal movement, and I'm incapable of getting excited about big tech companies owning another thing

But the other effect is going to be the massive loss in marginal value of any art form that gets a competitive AI. It takes humans weeks to years to write a book or a music album or a high quality painting, the AI is gonna generate all the music you could ever listen to for $10/month, and it'll do it in seconds. Wanna bet how many new bands are going to be able to launch careers against that? What company is going to hire an illustrator when ArtBot costs just $5000 a year for a commercial license*?

In a pure economic sense this is efficient, we get more art for less money, all while maximizing shareholder value. Its also going to deprive a lot of people the chance to do something creative that they love at an extremely high level. It'll probably keep a lot of people out of arts as a hobby as well; what's the point in learning to draw when ArtBot Basic does it better, immediately, and for free?

Seems a crying shame to me.

It's the same thing as happened to many other jobs over the last few centuries. There is still market for artisanal goods - just far smaller than before machines took over mass production. One should also look at simple physical work: machines changed it immensely, but it did not vanish - it just changed and includes ability to use forklifts and such. With artist AI it will be the same thing in my opinion: they still need an input from a user and getting out of it exactly what you want will require some skill and time. Time-efficiency will improve greatly and the required skills will change, but there will be jobs available.

Still, at some point we might end up in a situation, where there are not enough jobs for everyone. In a sense, we are already there, as consumption is artificially pumped to keep the economy going, but this is an entirely different discussion.

[gomipile]

Its also going to deprive a lot of people the chance to do something creative that they love at an extremely high level. It'll probably keep a lot of people out of arts as a hobby as well; what's the point in learning to draw when ArtBot Basic does it better, immediately, and for free?

Seems a crying shame to me.

To a certain extent, but perhaps not to the same degree or in the same way as you think.

Many licensed engineers working today only use a small fraction of the mathematics they learned in university at their jobs, and make heavy use of simulation software. But, as a society, we still value the fact that they spent four plus years learning how the mathematics is applied, and how to tell when a software model might be throwing out inappropriate results.

I don't think there will be a one-to-one analogy between that and what will happen with AI art. But, there is wiggle room in the possibilities of how this will all shake out over the coming decades.

[NichG]

I don't see any way they wouldn't do that, subscription services are where the money is at in software now, and at some point this is going to be a product.

Well again, OpenAI tried to go this route with DALL-E, but then someone essentially personally bankrolled training Stable Diffusion instead and distributed the weights for free to the public. So at least in this particular case, the opposite of what you're saying ended up happening.

Its not cheap to train one of these things, but between donations, crowd-sourced training, etc, its actually still possible. And the companies that are going all-in on 'just scale it big' are leaving a lot of optimizations on the table that the community picks up 6 months to a year later. That's why I can run Stable Diffusion on my PC, but DALL-E requires cluster-scale resources to do inference...

[Bohandas]

But the other effect is going to be the massive loss in marginal value of any art form that gets a competitive AI. It takes humans weeks to years to write a book or a music album or a high quality painting, the AI is gonna generate all the music you could ever listen to for $10/month, and it'll do it in seconds. Wanna bet how many new bands are going to be able to launch careers against that? What company is going to hire an illustrator when ArtBot costs just $5000 a year for a commercial license*?

There's no more lamplighters, milkmen, or town blacksmiths either

*ArtBot Inc retains sole copyright to all ArtBot output. A limited publication license is granted for ArtBot generated assets with subscription to ArtBot. Users must abide by the Terms of Service, which can change at any time. Upon termination of an ArtBot subscription, you lose all rights to ArtBot generated assets.

That;s really more of a problem with our copyright system allowing ai generated works to be copyrighted. Once writers and poets and artists are mechanized there's really no more reason for copyright to exist at all.

[georgie_leech]

I think the skills and resources needed to be an artist might change, but I do still think there will be a place for people to guide the AI's to produce interesting media. Like, the skills and equipment to compose music on a keyboard and computer might be drastically different than when you needed an orchestra to get the same sounds, but the underlying theory is the same. Outsourcing the theory bits to AI doesn't mean there's no more room for creative input or direction. I mean, heck, we give literal Directors a lot of credit for how movies come out, and lots of that work is done by other intelligences we call people

I think jokes aside though, the nature of copyright in a post-AI world is probably a little outside the scope and rules here, so we should probably step back from that particular angle...

[MetroAlien]

I just imagined a world where the majority of commercially used art is produced by AI.
In that world, all the AI was either trained long ago on the same dataset or, when trained new, most of the data came from other AI.

That's a huge crisis of creativity.
Now, to be fair, even without AI one of the biggest complaints people have about modern culture is that it's too derivative.

Come to think of it, major media corporations already function almost like the AI of my imagination, producing templated outputs, while indie creators are left in the dust.
In that way, being creative professionally already has its stumbling blocks.

but I disagree that people will stop being creative as a hobby.
If anything, new hobbies may emerge as a result.

I'm kinda baffled by the uncritical assumption that people necessarily need to be commercially competitive to engage in creative activities?

[NichG]

I just imagined a world where the majority of commercially used art is produced by AI.
In that world, all the AI was either trained long ago on the same dataset or, when trained new, most of the data came from other AI.

That's a huge crisis of creativity.
Now, to be fair, even without AI one of the biggest complaints people have about modern culture is that it's too derivative.

Come to think of it, major media corporations already function almost like the AI of my imagination, producing templated outputs, while indie creators are left in the dust.
In that way, being creative professionally already has its stumbling blocks.

but I disagree that people will stop being creative as a hobby.
If anything, new hobbies may emerge as a result.

I'm kinda baffled by the uncritical assumption that people necessarily need to be commercially competitive to engage in creative activities?

I mean, its a valid point that one consequence of this kind of thing is that its less possible to use certain enjoyable creative activities in order to also meet your survival needs. Which is a negative impact for the people who want to live that way or have invested resources in order to be able to live that way.

That said, I think the particular market that's likely to evaporate isn't actually the commercially used art, but rather the market for less-than-the-best art commissioned for personal use and indie-scale projects. Also probably edge stuff like mobile gacha games and things like that. The issue is that the bar for serious commercial usage is high enough that the occasional AI glitch like two-faced three-armed character portraits isn't really a tolerable stylistic choice for what the audience expects, so even if you want to use AI art you still are going to need to have a professional artist who also has technical experience with the AI system in order to get the needed outputs and fix the glitches by hand and so on.

But for someone making a hobby passion project tabletop RPG and wanting to illustrate it? Or someone making an indie game where the previous alternative to commissioning art was programmer art? Or someone who has $50 to spare to illustrate their character for a campaign they're super enthused about? Or someone writing a webnovel and thinking about going to Kindle Unlimited and they need cover art? Yeah, I would believe that those sources of income are likely to wane because of this technology.

That said, this is also possibly the very moment where someone who does have art skills and learns to integrate the AI stuff into their workflow can probably jump the queue and get into the professional art world far more readily than they could have two years ago (or two years from now).

So like a lot of things, opportunity is going to get shuffled around a lot, and that will suck for the people who are on the waning end and it'll be great for people who are on the waxing end.

As far as hobbyists though, this will definitely surge creative activity, just like any sort of barrier-lowering tools tend to do. Just think of stuff like RPGMaker...

[Bohandas]

I'm kinda baffled by the uncritical assumption that people necessarily need to be commercially competitive to engage in creative activities?

I'd go as far as to say that the two are actually mutually exclusive to a degree. If you're producing artwork for the man then you're every bit as much of an automaton as Artbreeder or Neural Blender

[MetroAlien]

The way I see art-AI being used in commercial settings, is that a company licenses an instance of a powerful AI, has an expert fine-tune it for their preference, and the project manager just keeps clicking "new image" until a good one shows up.

It neatly fits my mental image of a generic, lazy project manager solving everything with money rather than know-how.

[NichG]

The way I see art-AI being used in commercial settings, is that a company licenses an instance of a powerful AI, has an expert fine-tune it for their preference, and the project manager just keeps clicking "new image" until a good one shows up.

It neatly fits my mental image of a generic, lazy project manager solving everything with money rather than know-how.

The best AI artists out there are those who are very good at systematic curation. It's sort of the opposite of 'solving without know-how', since if you aren't good at curation you are going to ship things that don't hold up to scrutiny...

I think its probably true of traditional artists as well - those who lack the ability to self-critique accurately may well reach a lower plateau than those who are really good at noticing things that are wrong with their own work and correcting those things.

[MetroAlien]

in my imagination the suits don't know how to tell a "talented" AI from a mediocre one, and don't care enough to ask
They just think "AI innovative <=> good"

Like that one blood testing start up that turned out to be a scam.

I'm a pessimist, ok, I expect the worst.

[Tyndmyr]

There's no more lamplighters, milkmen, or town blacksmiths either

Exactly. Society changes. We won't need the same thing, but we'll need something new.

Imagine trying to explain jobs like "youtube influencer" to a lamplighter. There'd be a bit of a culture gap there, I think. In the same way, we can be sure that the future will be different than we have now, even if we cannot predict exactly how.

People will still do something, it just won't be the same things. Maybe the handling of a lot of basic art will result in artistry happening at a different level. Someone making a video game is art, and if some of the actual asset creation is handled by AI, that could be a boon to indie developers. Everything that gives the individual more ability to create is a good thing, I think. Not everyone will explore every capability with it, but you can be sure that some people will be creative, and open up something new.

[Bohandas]

I wonder if it would possible, by limiting resolution and color pallete, to make an art AI simple enough that a person with an afternoon to kill and a basic understanding of vector multiplication could solve the equations by hand. They could call it "Bot Ross"

edit: happy little decision trees

[NichG]

I wonder if it would possible, by limiting resolution and color pallete, to make an art AI simple enough that a person with an afternoon to kill and a basic understanding of vector multiplication could solve the equations by hand. They could call it "Bot Ross"

edit: happy decision trees

To get something that captures the qualities of natural images, probably not. To create a particular artistic image, maybe, but I doubt it...

Calculating an 8 x 8 image from a two layer CPPN with 8 hidden units might be doable. That's ~3000 operations. If you can do one sum or mult every 10 seconds, that would take you a bit less than 10 hours.

[gomipile]

To get something that captures the qualities of natural images, probably not. To create a particular artistic image, maybe, but I doubt it...

Calculating an 8 x 8 image from a two layer CPPN with 8 hidden units might be doable. That's ~3000 operations. If you can do one sum or mult every 10 seconds, that would take you a bit less than 10 hours.

Maybe if you could make it work with ReLU instead of sigmoid or Gaussian activation functions, it could cut down on the calculation complexity?

Rather than doing it by hand, I'd like to see what kinds of pretrained models could be run in reasonable time on an 80s 8-bit micro like a Sinclair, C64, Apple II, etc.

[NichG]

Maybe if you could make it work with ReLU instead of sigmoid or Gaussian activation functions, it could cut down on the calculation complexity?

Rather than doing it by hand, I'd like to see what kinds of pretrained models could be run in reasonable time on an 80s 8-bit micro like a Sinclair, C64, Apple II, etc.

I wasn't even counting those operations... this was just the matrix mults and bias additions...

[gomipile]

I wasn't even counting those operations... this was just the matrix mults and bias additions...

Well then. I'm reminded that there's a reason all those printed tables of logarithims, trigonometric functions and other special functions exist.

[Fat Rooster]

But for someone making a hobby passion project tabletop RPG and wanting to illustrate it? Or someone making an indie game where the previous alternative to commissioning art was programmer art? Or someone who has $50 to spare to illustrate their character for a campaign they're super enthused about? Or someone writing a webnovel and thinking about going to Kindle Unlimited and they need cover art? Yeah, I would believe that those sources of income are likely to wane because of this technology.

I'm not so sure about those specific examples. In particular the $50 to illustrate a character. The point is that this is a luxury item anyway, and they still have the $50 that they are going to spend on something. Instead of getting a single illustration they might be able to get a whole album based on stories from the campaign. The $100 they spend on gaming that would have gone $50 on a model and $50 on an illustration might become $40 on the model and $60 on a lot more artwork.

Cover images are marketing that is always going to be worth spending money on even marginal improvements, so I don't see that market disappearing either.

The indie game dev who is now doing their own art? That's harder to judge the outcomes of, but a lower barrier for minimal viable product is probably a good thing. They will still be using programmer art. It is just that that phrase will mean something slightly different.

The point is that being able to get more for less doesn't mean revenues will fall for products that are not strictly necessary anyway. The question is not "how many artworks does the world need", it is closer to "how much art can we get for $X". It is entirely possible that more art for the same money will mean more money is diverted to artists, not less.

It is definitely going to be messy, and artists who are not able or willing to embrace the new tools are probably going to struggle. I am heartened by memories of an episode of 'faking it' though, where they took a classically trained musician and had them train for a week to fake being a DJ. She absolutely nailed it, because she was extremely musically gifted. The best AI assistant is not going to make me an artist. I think the overwhelming majority of artists will take to the tools easily and be able maintain their living close to or above where it is now, because art will still be something people are prepared to spend money on, and they are better at it.

[NichG]

I'm not so sure about those specific examples. In particular the $50 to illustrate a character. The point is that this is a luxury item anyway, and they still have the $50 that they are going to spend on something. Instead of getting a single illustration they might be able to get a whole album based on stories from the campaign. The $100 they spend on gaming that would have gone $50 on a model and $50 on an illustration might become $40 on the model and $60 on a lot more artwork.

Cover images are marketing that is always going to be worth spending money on even marginal improvements, so I don't see that market disappearing either.

The indie game dev who is now doing their own art? That's harder to judge the outcomes of, but a lower barrier for minimal viable product is probably a good thing. They will still be using programmer art. It is just that that phrase will mean something slightly different.

The point is that being able to get more for less doesn't mean revenues will fall for products that are not strictly necessary anyway. The question is not "how many artworks does the world need", it is closer to "how much art can we get for $X". It is entirely possible that more art for the same money will mean more money is diverted to artists, not less.

It is definitely going to be messy, and artists who are not able or willing to embrace the new tools are probably going to struggle. I am heartened by memories of an episode of 'faking it' though, where they took a classically trained musician and had them train for a week to fake being a DJ. She absolutely nailed it, because she was extremely musically gifted. The best AI assistant is not going to make me an artist. I think the overwhelming majority of artists will take to the tools easily and be able maintain their living close to or above where it is now, because art will still be something people are prepared to spend money on, and they are better at it.

Not so much a matter of getting more for less, more a matter of who gets what. Instead of that going to someone trying to build a portfolio for commercial work, it will probably go to AWS or Google Cloud or to NVidia for a higher end GPU.

I was paying something like $200 a pop for concept art style backgrounds for Travelogue when I was thinking of releasing it on Steam. Doing the inventory icons would have cost close to $10k if we had actually found someone to do it before the project lost momentum. If I were doing it again I'd just do all of that in Stable Diffusion over the course of a day or two now - no question.

[Eldan]

Not so much a matter of getting more for less, more a matter of who gets what. Instead of that going to someone trying to build a portfolio for commercial work, it will probably go to AWS or Google Cloud or to NVidia for a higher end GPU.

I was paying something like $200 a pop for concept art style backgrounds for Travelogue when I was thinking of releasing it on Steam. Doing the inventory icons would have cost close to $10k if we had actually found someone to do it before the project lost momentum. If I were doing it again I'd just do all of that in Stable Diffusion over the course of a day or two now - no question.

Yeah, same. I'm working on some setting books with some other people. We have hundreds of pages written, but we don't really want to release without artwork. We got some donated art, so that we could put out a few books already (with less artwork than we'd like), but our publishing speed suddenly quadrupled when four or five of us got Midjourney accounts. We can now just do things like page number icons, or fancy scrollwork around titles ourselves in a few hours.

[137beth]

One webcomic I enjoyed that ended way too early is Galebound. The comic covered only about 1/5 of the story the author had planned before she gave up on doing it as a comic. She felt that it was taking too long to draw, and she prefers writing over drawing anyways, so she has rewritten it as a novel and is working on publishing it in some form.

In April of this year, she posted an Artbreeder-generated image of one of the main characters from Galebound on Discord. While the visual style Artbreeder produced is very different from how she drew the comic, it was still recognizably a picture of her character.

I still expect her to eventually release Galebound as a text-only novel. However, as AI-generated art becomes more versatile and accessible, I would not be surprised if she goes back and redoes the comic with the assistance of AI art.

[1dominator]

I'm really into it. I have a paid account on Midjourney now, for unlimited art generation, after one of my roleplaying discord channels started to heavily use the artwork for monsters, landscapes and NPCs and we were really impressed.

Here's some of the stuff we've made and are using in game. These are all real-time generated by a discord bot. (About a minute per picture.)

Places:

Spoiler

Show

Cyberpunk city:

I really like this "cyber punk city" one. I don't suppose you could upload a higher res version of it?

[gomipile]

I really like this "cyber punk city" one. I don't suppose you could upload a higher res version of it?

Given the thread topic, I suppose using an AI upscaler would be the on-topic choice.

[Eldan]

Sure, here you go. Got some fast time left this month, so I can upscale.

Spoiler

Show

And a lower detail version, which I think often looks better:

Spoiler

Show

I have a stupidly huge gallery by now, since I've been doing this most days for months now, but feel free to browse.

https://www.midjourney.com/app/users...3937003249677/

[1dominator]

Sure, here you go. Got some fast time left this month, so I can upscale.

Spoiler

Show

And a lower detail version, which I think often looks better:

Spoiler

Show

I have a stupidly huge gallery by now, since I've been doing this most days for months now, but feel free to browse.

https://www.midjourney.com/app/users...3937003249677/

Thank you, they're both great.

[Bohandas]

It looks more like Barad-Dur to me

[Eldan]

Yes, but with electric lights. My RPG group calls it Barad-Yutani. (I know it's the wrong corp, but Tyrell-Dur didn't work).