Why don't indie devs use AI-generated images as art?

[V17]

Imo anyone who says that "real artists" don't use shit like AI generated art has no idea how most commercial graphic artists work [,,,,] The reason why it's not used much yet

I agree with what you're saying about how artists generally work here (it's the same with music, and actually I think all professions, even the law), but aren't you contradicting yourself above? It would make more sense if instead of "don't" you said "wouldn't." Sorry for the tism

I was trying to speak in general, about all possible sources of inspiration that already exist (like say Pinterest - I fucking hate it, but it works), but I'm also not a native speaker and it was late at night, so you're probably right.

 

[J1M]

Imagine the decline of our society when AS ( artificial stupidity) algorythm produced images are on par or better than human ( lack of) creativity .

Not gonna happen since AI needs human-made references to train, so at very best it will be on par with the average of what it has been fed.

The training data doesn't need to be created by humans, only curated by them.

 

[Hag]

The training data doesn't need to be created by humans, only curated by them.

Same issue. The AI doesn't know what it is doing. It's braindead. It is weighted randomness. It is entirely dependent on human interference, either to provide the initial data, to curate parameters, to select the results or simply to improve the algorithm.

Self-driving cars failed because people are trained to drive and it happens that the average Joe is a better driver than the best effort of some of the most powerful companies of the world.
However, most people are not trained on fine arts and don't give a shit about it (the fact that we are surrounded by trash art everywhere doesn't help), so indeed AI generated pictures are good enough for them. It doesn't mean it gets any close to what a moderately competent artist can do.

The fact that some AI generated art looks remotely like what a modern artist can do says more about the sorry state of art than the prowess of technology.

 

[V17]

Imagine the decline of our society when AS ( artificial stupidity) algorythm produced images are on par or better than human ( lack of) creativity .

Not gonna happen since AI needs human-made references to train, so at very best it will be on par with the average of what it has been fed.

If somebody curates the results, it doesn't need to be the average. It just won't be better and significantly more innovative than the training datasets. And I would argue anything more is not necessary for videogames. Actually good art that pushes the visual medium forward generally doesn't happen there anyway, never has. There's no modern Rembrandts even in the best videogames.

 

[J1M]

Games shouldn't be trying to invent a new art style. That's like trying to invent a new franchise by ordering a million printed lunch boxes.

Game art should focus on supporting the player experience, such as being instantly clear at conveying information and supporting the fantasy/role of the player.

 

[Üstad]

AI generated art is important because it's less time consuming and cost effective. Artists are limited and so their time. I don't think AI will beat humans in art anywhere soon but it has good commercial use.

This image has been created merely in a few minutes, obviously it'll get better and people will make use of the mass production.

 

[rusty_shackleford]

Self-driving cars failed

They didn't fail because they've never even been tested. The main issue is the people constantly pushing the "SELF DRIVING CARS WILL BE HERE IN 5 10 15 20 25 30 SOMETIME SOON YEARS!!!!". These things, they take time.

 

[Hag]

They didn't fail because they've never even been tested. The main issue is the people constantly pushing the "SELF DRIVING CARS WILL BE HERE IN 5 10 15 20 25 30 SOMETIME SOON YEARS!!!!". These things, they take time.

Must have been over twenty years since I've first heard about self-driving cars trials. They do take time indeed and I'm sure I'll never see them in my lifetime, not without a brand new technological approach.

 

[J1M]

They didn't fail because they've never even been tested. The main issue is the people constantly pushing the "SELF DRIVING CARS WILL BE HERE IN 5 10 15 20 25 30 SOMETIME SOON YEARS!!!!". These things, they take time.

Must have been over twenty years since I've first heard about self-driving cars trials. They do take time indeed and I'm sure I'll never see them in my lifetime, not without a brand new technological approach.

There are too many Americans that drive for a living. There needs to be a shift in the workforce before it could be accepted without killing the economy. I expect things like drone delivery to happen first. But also any real plan for self-driving cars needs to demonstrate some replacement jobs to pass legislation requirements.

 

[gurugeorge]

They didn't fail because they've never even been tested. The main issue is the people constantly pushing the "SELF DRIVING CARS WILL BE HERE IN 5 10 15 20 25 30 SOMETIME SOON YEARS!!!!". These things, they take time.

Must have been over twenty years since I've first heard about self-driving cars trials. They do take time indeed and I'm sure I'll never see them in my lifetime, not without a brand new technological approach.

Eh, it's the same with the whole AI thing. It's been "just around the corner" every 10 years or so for decades.

Part of the problem may be that intelligence has something deeply to do with sociality, so that the "mind" isn't just the thing in the brain-case, but patterns of activity and reactivity that are unconsciously shared across a group of things-in-braincases. Rather analogously to the way the market is a giant, primitive AI. Also probably related to the Hegelian/Marxist intuition about the systemic nature of society (how the larger patterns of social interaction affect individuals). So there's probably still a way to go to get AI - because you're having to match not just the complexity of one brain, but the complexity of millions of brains working on problems with shared signals. For example, most words (nouns) implicitly "contain" potential action-patterns - e.g. "bird" contains the expectation of certain results - typical bird-like behaviours - consequent on certain interactions with "it".

Stuff like that.

 

[agris]

A big issue is just usability. You can get some really interesting designs and stills - but apart from using it as some elements in a game, its not really 'there' yet. Like - for background loading scenes perhaps, and character portraits - but if you want to take the step where you have a game character, sprite, animation, etc - AI art isn't useful. It's basically stock photographs. You could make a game if you had access to stock photography - but only very specific games could be made.

To the people it IS already useful to, they have artistic skill nessesary to USE it in interesting ways. But it's not going to replace game artists, or really help non artistic developers create usable game assets in its current form.

are you aware of any AI art gen that will produce *.AI/PSD/EPS/XCF with different design elements preserved as individual layers? Giving artists access to the layers would go a long way to increasing the usability of these machine generated images, but I imagine it's a huge computational task and the technology isn't there yet.

 

[The_Sloth_Sleeps]

Oh look! I can use it to generate the art for my isometric game. This will really speed up production.

 

[V17]

They didn't fail because they've never even been tested. The main issue is the people constantly pushing the "SELF DRIVING CARS WILL BE HERE IN 5 10 15 20 25 30 SOMETIME SOON YEARS!!!!". These things, they take time.

Must have been over twenty years since I've first heard about self-driving cars trials. They do take time indeed and I'm sure I'll never see them in my lifetime, not without a brand new technological approach.

Current neural networks basically are brand new. They're a very old concept, but large scale GPU computation of neural networks that revolutionized their research only became relatively cheap in 2010s. We're still in the "low hanging fruit" phase of research where small breakthroughs happen pretty regularly. Judging self-driving cars (or any other current neural net applications) on that just because Musk has a big mouth does not make any sense.

AIs usable for game-ready graphics will undoubtedly happen too at some point, it's just a question if there's enough money in it for somebody to do the research and development sometime soon of if we're going to have to wait till somebody else does most of the work on the tech for creating similar but more profitable stuff. Imo even the usability of neural nets for concept art workflow is a huge thing on its own.

Also the tech evolves really fast. Artflow referenced above was launched a year ago and is quite limited, afaik they specialize on portraits and landscapes (which indeed makes it a big surprise they can't make an isometric house), where as Craiyon I think launched like two months ago and is far from state of the art, but it at least already understands what an isometric house is:

 

[Bigg Boss]

Because it's not art.

Cope
Nobody cares if it's not made by humans except for art*sts put out of a job.

This is the truth of it. I know so many artists (that don't work) that whine about this stuff. They are butthurt normies are making pretty art but not really making it while they are also not making art. Hilarious.

I did one for fun.

I am so talented.

 

[J1M]

People get upset when their job security is threatened. Can't see that they could easily make a cool million contracting out at current prices using new tools before the rest of their peers adopt them.

 

[Haba]

People get upset when their job security is threatened. Can't see that they could easily make a cool million contracting out at current prices using new tools before the rest of their peers adopt them.

"I like that picture, could you make the character female?"
"Uh..."
"Don't you have a higher resolution image?"
"Um..."

 

[Reinhardt]

AI is raycis and will erase black people of afro-american color from the games!

 

[The_Sloth_Sleeps]

Some of the previous comments is what I have grown to hate about academia. The loudest shouting liars get all the attention.

-If there's a screenshot that happens to look randomly ok, it will get put in the paper, not the 99.9% shyte looking ones.
-Generated videos/games. They will only show the runs that don't look like utter nonsense. They pretend they are much further along than they are.

Honestly all this stuff is such a steaming pile of Garbage. I really don't see where any of the supposed advancements are that are not to do with improved computing power/more data. It all looks as bad as its ever been. Technology is not improving.

 

[V17]

Some of the previous comments is what I have grown to hate about academia. The loudest shouting liars get all the attention.

-If there's a screenshot that happens to look randomly ok, it will get put in the paper, not the 99.9% shyte looking ones.
-Generated videos/games. They will only show the runs that don't look like utter nonsense. They pretend they are much further along than they are.

Honestly all this stuff is such a steaming pile of Garbage. I really don't see where any of the supposed advancements are that are not to do with improved computing power/more data. It all looks as bad as its ever been. Technology is not improving.

I am starting to be convinced you cannot possibly be this retarded and I'm feeding a troll, but just in case: nobody professional is using this free tool that was made for fun (I only used is a counterexample to your stupid isometric house attempt), nobody sane is claiming that AI generates art that is final or useful without curation. I'd send you the actual concept art work made using Dall-E and Midjourney, but it's commercial work that was shared with me in a private forum, so I cannot publicly redistribute it. It's just a matter of time till examples of such workflows are shown in youtube videos etc. though.

 

[The_Sloth_Sleeps]

They didn't fail because they've never even been tested. The main issue is the people constantly pushing the "SELF DRIVING CARS WILL BE HERE IN 5 10 15 20 25 30 SOMETIME SOON YEARS!!!!". These things, they take time.

Must have been over twenty years since I've first heard about self-driving cars trials. They do take time indeed and I'm sure I'll never see them in my lifetime, not without a brand new technological approach.

Current neural networks basically are brand new. They're a very old concept, but large scale GPU computation of neural networks that revolutionized their research only became relatively cheap in 2010s. We're still in the "low hanging fruit" phase of research where small breakthroughs happen pretty regularly. Judging self-driving cars (or any other current neural net applications) on that just because Musk has a big mouth does not make any sense.

AIs usable for game-ready graphics will undoubtedly happen too at some point, it's just a question if there's enough money in it for somebody to do the research and development sometime soon of if we're going to have to wait till somebody else does most of the work on the tech for creating similar but more profitable stuff. Imo even the usability of neural nets for concept art workflow is a huge thing on its own.

Also the tech evolves really fast. Artflow referenced above was launched a year ago and is quite limited, afaik they specialize on portraits and landscapes (which indeed makes it a big surprise they can't make an isometric house), where as Craiyon I think launched like two months ago and is far from state of the art, but it at least already understands what an isometric house is:

This result is less useful than a Google image search. (Read the title of the thread).

Of course the generated art is rubbish and no use with hiring an artist on top of this (which begs whats the point?).

My thought is: are there *any* use for this stuff at all? I have my doubts that for anything this algorithm produces there isn't already an existing algorithm that outperforms it.

 

[V17]

They didn't fail because they've never even been tested. The main issue is the people constantly pushing the "SELF DRIVING CARS WILL BE HERE IN 5 10 15 20 25 30 SOMETIME SOON YEARS!!!!". These things, they take time.

Must have been over twenty years since I've first heard about self-driving cars trials. They do take time indeed and I'm sure I'll never see them in my lifetime, not without a brand new technological approach.

Current neural networks basically are brand new. They're a very old concept, but large scale GPU computation of neural networks that revolutionized their research only became relatively cheap in 2010s. We're still in the "low hanging fruit" phase of research where small breakthroughs happen pretty regularly. Judging self-driving cars (or any other current neural net applications) on that just because Musk has a big mouth does not make any sense.

AIs usable for game-ready graphics will undoubtedly happen too at some point, it's just a question if there's enough money in it for somebody to do the research and development sometime soon of if we're going to have to wait till somebody else does most of the work on the tech for creating similar but more profitable stuff. Imo even the usability of neural nets for concept art workflow is a huge thing on its own.

Also the tech evolves really fast. Artflow referenced above was launched a year ago and is quite limited, afaik they specialize on portraits and landscapes (which indeed makes it a big surprise they can't make an isometric house), where as Craiyon I think launched like two months ago and is far from state of the art, but it at least already understands what an isometric house is:

This result is less useful than a Google image search. (Read the title of the thread).

Of course the generated art is rubbish and no use with hiring an artist on top of this (which begs whats the point?).

My thought is: are there *any* use for this stuff at all? I have my doubts that for anything this algorithm produces there isn't already an existing algorithm that outperforms it.

You've already been told several times, but let's try it again. When creating game art (or vfx or achviz or product viz or...) there's generally a concept art stage, which is where most of the creative work needs to be done. That creativity doesn't come from thin air, there are processes that artists use. And AIs like Midjourney can make that workflow significantly faster because with a skilled user it's a powerful and fast inspiration machine. In some cases its output can be used straight up just with some editing, in other cases it's just used as a base for something else. The AI is designed for this type of work and allows for things like uploading your own image into it and having it create several variations, recreating it in a different style, in higher resolutions etc. Don't expect people to make examples just for you because these services naturally cost money to use.

You're talking theoretically about something you don't understand. The reality is that the technology is already being used and when in a few years you see the graphics made by artists who used it, you won't even know about it.

 

[J1M]

People get upset when their job security is threatened. Can't see that they could easily make a cool million contracting out at current prices using new tools before the rest of their peers adopt them.

"I like that picture, could you make the character female?"
"Uh..."
"Don't you have a higher resolution image?"
"Um..."

Don't worry, in 5 years there will be a job that will train you on the workflow.

 

[Hag]

You've already been told several times, but let's try it again. When creating game art (or vfx or achviz or product viz or...) there's generally a concept art stage, which is where most of the creative work needs to be done. That creativity doesn't come from thin air, there are processes that artists use. And AIs like Midjourney can make that workflow significantly faster because with a skilled user it's a powerful and fast inspiration machine. In some cases its output can be used straight up just with some editing, in other cases it's just used as a base for something else. The AI is designed for this type of work and allows for things like uploading your own image into it and having it create several variations, recreating it in a different style, in higher resolutions etc. Don't expect people to make examples just for you because these services naturally cost money to use.

You're talking theoretically about something you don't understand. The reality is that the technology is already being used and when in a few years you see the graphics made by artists who used it, you won't even know about it.

Once again you're not talking about quality, you're not talking about coherence, about training artists to be better, about making art people enjoy, nor about exceeding expectations, nor anything that makes art beautiful or touching. You're not interested on the fact that creative vision can transform a game, or that creativity is often seen as the most pleasant part of concepts design.

You're talking theoretically about something you don't understand. You don't understand AI. You don't see that most art and design is already in a quagmire, that mediocrity and uniformity have taken over all creative endeavor. Go on, boast about the fact that AI lets you shit designs quicker than an amoebia-ridden colon, rejoice in the great look-alike CGI pictures carpet-bombing, revel in your own inability to experience critical thinking and beauty appreciation. You're a bean counter. You want to optimize workflows. You care about money, not people, not even results, you're boring and bring nothing of value, you talk with internet keywords and does not seem to show any specie of original thoughts. You could be replaced by some AI.

 

[The_Sloth_Sleeps]

They didn't fail because they've never even been tested. The main issue is the people constantly pushing the "SELF DRIVING CARS WILL BE HERE IN 5 10 15 20 25 30 SOMETIME SOON YEARS!!!!". These things, they take time.

Must have been over twenty years since I've first heard about self-driving cars trials. They do take time indeed and I'm sure I'll never see them in my lifetime, not without a brand new technological approach.

Current neural networks basically are brand new. They're a very old concept, but large scale GPU computation of neural networks that revolutionized their research only became relatively cheap in 2010s. We're still in the "low hanging fruit" phase of research where small breakthroughs happen pretty regularly. Judging self-driving cars (or any other current neural net applications) on that just because Musk has a big mouth does not make any sense.

AIs usable for game-ready graphics will undoubtedly happen too at some point, it's just a question if there's enough money in it for somebody to do the research and development sometime soon of if we're going to have to wait till somebody else does most of the work on the tech for creating similar but more profitable stuff. Imo even the usability of neural nets for concept art workflow is a huge thing on its own.

Also the tech evolves really fast. Artflow referenced above was launched a year ago and is quite limited, afaik they specialize on portraits and landscapes (which indeed makes it a big surprise they can't make an isometric house), where as Craiyon I think launched like two months ago and is far from state of the art, but it at least already understands what an isometric house is:

This result is less useful than a Google image search. (Read the title of the thread).

Of course the generated art is rubbish and no use with hiring an artist on top of this (which begs whats the point?).

My thought is: are there *any* use for this stuff at all? I have my doubts that for anything this algorithm produces there isn't already an existing algorithm that outperforms it.

You've already been told several times, but let's try it again. When creating game art (or vfx or achviz or product viz or...) there's generally a concept art stage, which is where most of the creative work needs to be done. That creativity doesn't come from thin air, there are processes that artists use. And AIs like Midjourney can make that workflow significantly faster because with a skilled user it's a powerful and fast inspiration machine. In some cases its output can be used straight up just with some editing, in other cases it's just used as a base for something else. The AI is designed for this type of work and allows for things like uploading your own image into it and having it create several variations, recreating it in a different style, in higher resolutions etc. Don't expect people to make examples just for you because these services naturally cost money to use.

You're talking theoretically about something you don't understand. The reality is that the technology is already being used and when in a few years you see the graphics made by artists who used it, you won't even know about it.

Either take a shit or get off the toilet. A person can draw inspiration from anything, so basically its the people doing all the difficult stuff and your "AI" isn't bringing much to the table. I'm still struggling to see the use, and theres no good examples (that aren't fluke one in a million results). All the online apps I've tried have toilet results.

As for more practical uses for someone like me, a developer who has to pay for artists, this retarded algorithm can't even generate a decent face or isometric building, how on earth could I use this?

 

[gurugeorge]

Once again you're not talking about quality, you're not talking about coherence, about training artists to be better, about making art people enjoy, nor about exceeding expectations, nor anything that makes art beautiful or touching.

You're both right actually, V17 is right about the aspect of it that's a job, that has targets, and how professionals actually get things done, etc. And you're right about the decline in standards.

But the decline in standards is deliberately engineered from the top, it's not a function of the tools artists use, or the way they improve their workflows, etc., those working principles would hold even in a context with a more elevated level of quality in art design.