Your thoughts on A.I. art creation

[KyriakosCH]

^Regarding AI establishing what works for humans as good art, I have to suppose this would quickly approach one of the many (already) known limitations of any level of a formal logic system. Of course the program can produce stuff and 'by accident' achieve very high quality, but I doubt it will be able itself to reproduce that (more likely that it will mathematically model the success as tied to parameters of the image it wasn't tied to for the human), and even if it does, it won't go above that level, while a human can (due to serendipity if not conscious calculation and extrapolation).

What I mean, the question is whether the human thought and perception may be described mathematically. So far most of the processes in human body were, at least this is my impression based on what I've read or heard. There's already some understanding of how human brain works. I cannot predict the future, but I think there's a realistic chance that, given time and effort, humans will be able to define how their own mind works, including intuition, perception of beauty, and so forth.

Of course the above assumes that the nature of our thought is deterministic. If there's, say, a spiritual essence which cannot be described by a formal logic, then we have a different situation...

I was only alluding to formal logic systems having inherent blind-spots (eg Goedel sentences), which aren't removed regardless of expansions of the system.
But I won't be at all surprised if the human mind (despite being closer to an analog machine, or at least looking like such) also has built-in defenses against everything being provable (or understandable) while still maintaining consistency.
Besides, even philosophically, how could one be aware of what being aware means, without that altering what being aware means? =>some level outside the stable one would appear to be needed, so the unawareness is just pushed there.

[Crimson Wizard]

I was only alluding to formal logic systems having inherent blind-spots (eg Goedel sentences), which aren't removed regardless of expansions of the system.

I'm not going to lie that I understand this fully; it's been many years since I studied math in any way.

But something that I might say: the systems that produce complicated results are not necessarily based on complicated rules. Sometimes complex systems are complex because they are vast, and have many elements; but at the same time may have a simple foundation, which is then copied many times, with variations, or different input parameters.

If the mind has similar structure, then it's a matter of learning the foundations, after which calculating the thought, so to speak, would be rather an issue of employing enough calculation power rather than devising a difficult mathematical theorem.

But I won't be at all surprised if the human mind (despite being closer to an analog machine, or at least looking like such) also has built-in defenses against everything being provable (or understandable) while still maintaining consistency.

Well, in that case, I guess, the solution is in building an AI that can understand us. Which may be a good joke, but also a truth.

[Babar]

Bad ones do...

A bad artist using shortcuts is never going to surpass a good artist using shortcuts. In fact, they're probably not even going to surpass a good artist not using shortcuts.

But you can't argue both points; either AI art is good enough to provide a useful product (proto-art that a human can then expand upon, or use in creating a finished piece of art), or AI art is garbage that nobody in their right mind would try to pass off as acceptable. If the AI is making bad art, why would an artist want to use that as a base to work from? On the other hand, if AI will eventually become "better", why limit it, why not let it take over content generation completely?

I don't think my point is contradictory- something being usable as a base or source or inspiration is different from something being usable as is. A good artist can use it to save time in their process of making original art, a bad artist can use it as is, and be called out for having horrible art (or find a use case where quality doesn't matter).
 

There's no doubt, that in some form, AI could be used effectively as another tool in making artwork (much like Photoshop), whether with advanced compositing, background removal, etc*. What I object to is the concept of "creative" AI; generating content alone, not with creativity or insight, but by algorithmic necromancy. Quality is irrelevant.

You put it in quotes, so you understand already, but it is important to hammer this point in- there is no such thing as "creative" AI. The way AI has been trained is by looking at thousands of images so that when you give it a fill in the blanks art question, it fills in the blank with the answer that has the highest score according to its algorithm. If we are talking about a potential future where AI is making art all on its own and it is something purely creative...we're nowhere near there yet.

As I said in my earlier post, that indeed is the end result sought by those at the top of the food chain. Sensible people like you and me might see it as folly, but that's where we're heading.

In the situations where such things are in focus, it doesn't matter what the people at the top of the food chain want. If they try to implement a process for their products purely using AI, their quality will suffer, and people will move to other products. If they find a niche where the actual quality doesn't matter as much and people get that, then...ok. It still won't take away from people who actually put in effort into their art. The example you linked above was AI generated children's books, and children not having that good a grasp of quality, it sounds like it would be an exploitable niche, except even then, the guy has only earned $100, and I'm pretty sure he's not going to eclipse any actually skilled writers.

[Snarky]

Over one hundred years of fabulous book covers (or record covers, movie posters...even toy boxes!) show that you don't have to reduce commercial promotional art to the status of the purely functional. Oh, you can, and yes, I'm certain that publishers/manufacturers would indeed be welcoming of purely AI-generated content in this sense.

And I'm sure that there will still be fabulous book covers, record covers, movie posters and other visual designs made. What AI-generated art will mainly replace (in part) is the 90% of book covers (etc.) that are just stock photos mashed together in Photoshop. (Another article in Norwegian.)

And the thing is: a great book cover can be nice in itself, but it doesn't actually affect the quality of the book. Some of my favorite books have absolutely god-awful covers (since the first edition I don't think The Fifth Head of Cerberus has ever been republished with a good cover; this one is particularly hideous), while a book you pick up because of a stylish cover can often be a disappointing read. Someone should come up with a saying about that...

And why deny the guy who designs your milk carton a job, simply because nobody cares what a milk carton looks like as long as it has milk in it?

I agree, that's a shame. I'm not sure it's more of a shame than anyone else losing their job to new technology, but in general I absolutely agree that there is a big danger that AI will eliminate a lot of jobs very quickly, and that people will suffer for it.

But don't you see that as a diminishing of the art, regardless of it's status? Maurice Noble created some astonishing backgrounds for old Warner Bros. cartoons (backgrounds that sometimes whizzed by in the blink of an eye during a Road Runner short), and I'm sure, as he cashed his weekly check, he didn't consider his work would end up as museum pieces. Or what about the beautifully designed backgrounds in Samurai Jack? People care. Granted, not everybody cares, but still. To argue for lesser art, simply because of it's nature, confuses me.

But there is a flipside to that, because lowering the barriers to entry will allow more people to make art. I know you pooh-poohed that before, saying that the thing keeping them back is actually lack of talent, but you've got to consider the people who have a great idea for a computer game but can't make graphics, or the people with a great idea for a cartoon who don't have the time or money to draw everything, etc., etc. Because like I said, a lot of art is just in service of some other/greater work, and merely because someone doesn't have the skill (or the time, or the money, or the friends/connections) to do that particular part, it doesn't mean they don't have artistic talent.

And sure, there will be a lot of crap. Just like AGS, by making it a lot easier to make adventure games, has led to the creation of a lot of terrible adventure games.

[KyriakosCH]

Personally I am not seeing a positive reason to be infatuated with the technical part of (say) creating visual art. Yes, it takes time to form such a skill. But the skill itself rarely (at least consciously) directs the more purely creative part of thinking of the work. It is more of a barrier, in that if your skill is insufficient, you won't be able to fully recreate what you had in your head, despite that already being fully formed in your mind.
It's also, itself, not apparently tied to other insights. For example, take someone who creates a 3d model, using a program like Blender or 3dsmax etc, and renders that as 2d. And theoretically assume someone else would have been able (not very likely, but not strictly impossible either) to produce the same 2d image, using a web of mathematical functions (which, ultimately, the 3d modeller makes use of itself, but on a level you don't typically access as a modeler). The latter would have insight on those functions too, while the former operates on a far looser connection to forms, associating them with the means to something and not themselves-for-themselves tied to insights.
It's the same in painting; you need time to form the skills needed to draw by hand or apply color, but those are just means to an end, not something which typically (let alone consciously) affects your creativity; it just stands as a barrier to achieving the potential.

Likewise with stuff of the ags variety: you have to learn commands and the structure used specifically here, but I doubt anyone would seriously claim it is the ags-skill which matters and not the result as a game. In that regard, how could it ever be not positive to be in a position to simply have the result you imagined, by not needing to type any command? (though, of course, with current tech this isn't possible, it's fairly conceivable that in the future it will be). 

[AndreasBlack]

My first attempt at AI usage. Bg Girl vocal. A bit Noisy, but it works in the mix i think. I'm not gonna lie. I like AI vocals rather then having to deal with problematic people when i can sing it myself in falsetto and change it to a girl https://soundcloud.com/maxfury_official/max-fury-girl-of-my-dreams

[eri0o]

https://civitai.com/models/151539/lucasarts-style-1990s-pc-adventure-games-sdxl-lora-dreambooth-trained

Saw this recently - I have alerts for Adventure Game Studio in search engines that I occasionally get.

Btw from the examples it doesn't look like it is useful beyond static images, and it's all very lifeless, but I could see perhaps using it for some text game where you navigate a place and polishing whatever is generated later in Aseprite by hand - adjusting colors, palette, resolution, clean up of the hallucinations, ... Still could be a really fast pipeline if you are working in intense time restrictions - like Ludum Dare, where you must work solo and output something in 48hours.

[cat]

Since AI learns from texts and pictures and then again creates text and pictures, I wonder when the moment will happen when AIs train mostly on AI generated stuff. What will this result in?
I just have to think of the Habsburgs, interbreeding for generations with a questionable outcome...

[Danvzare]

Since AI learns from texts and pictures and then again creates text and pictures, I wonder when the moment will happen when AIs train mostly on AI generated stuff. What will this result in?
I just have to think of the Habsburgs, interbreeding for generations with a questionable outcome...

I think we're all wondering that. Presumably it won't be as catastrophic as one might first think, because the text and pictures generated by AI that make their way online, tend to be curated by actual humans. For example, someone generates a hundred pictures using AI and then uploads the best one.

Still, what do you get if you keep feeding AI it's own output?
...
Ugh, I think I just answered my own question. That sounds like what they did to that two headed goat on Futurama. 

[AndreasBlack]

https://civitai.com/models/151539/lucasarts-style-1990s-pc-adventure-games-sdxl-lora-dreambooth-trained

Saw this recently - I have alerts for Adventure Game Studio in search engines that I occasionally get.

I have to agree. As for the voice AI however it's incredible if the sound source is really good!   I've done my own voice AI's and friend's AI voice profiles and used that in my game so far. It sounds awesome! It takes time for sure, but that's to expect. But it's doable if you really want voice acting and just can voice act yourself and then replace your voice with various profiles! I've done 400 lines atm from Female to male voices and most of them are good enough, i will probably go over some of them in the future and re-act, but it's good for now! And then if there's a line i can't do myself i just ask the friends to act it out. That's how i've done it so far

[eri0o]

Found a small thread here that doesn't have much things discussed, but posed an interesting idea I had not thought about AI: using AI for testing games. Of course, the subject matter is specific for Chat GPT and Parser games.

https://intfiction.org/t/testing-your-game-with-chatgpt/59863


Of course, perhaps just some automation that clicks in a gazillion random places in AGS and types random bs in a keyboard if stuck for too long in the same place could potentially uncover bugs/crashes with a much simpler code - no AI required.

[gijsja]

Testdrive this GPTBuild for AGS and tell me what you think
https://chat.openai.com/g/g-QGodtBnLc-adventure-pixel-director

[Babar]

And also not pixel art 

[Matti]

Since AI learns from texts and pictures and then again creates text and pictures, I wonder when the moment will happen when AIs train mostly on AI generated stuff. What will this result in?
I just have to think of the Habsburgs, interbreeding for generations with a questionable outcome...

I guess it will become normal that people have an arbitrary number of limbs and fingers 

[RootBound]

This article presents some pretty damning results. The further you get into their investigation, the worse it gets.

https://spectrum.ieee.org/midjourney-copyright

[LimpingFish]

Of course it plagiarizes. It can't create, it can only imitate. And in the six months since I last posted in this thread, the situation has only got scummier and the foaming greed around AI more apparent.

So call me a Luddite, but I'll say it again: Fuck Generative AI, and double fuck the cretins shilling it.

[cat]

Sometimes I print coloring pages for my daughter. Recently, she requested "a zebra that is jumping on a cloud". I told her, that this will not be possible and she was really sad.
However, yesterday I played around with Bing a bit and its image generation (it uses DALL-E). There I was able to create a lovely picture of a cute zebra jumping on a cloud in a style suitable as coloring page. My daughter was very happy (even if one of the zebras had five legs).

[Mandle]

Of course it plagiarizes. It can't create, it can only imitate. And in the six months since I last posted in this thread, the situation has only got scummier and the foaming greed around AI more apparent.

So call me a Luddite, but I'll say it again: Fuck Generative AI, and double fuck the cretins shilling it.

I tend to agree. It's interesting to watch how technology is outstripping our ability to keep up with it as a society. This was always gonna be the endgame though, I guess. There's no stopping progress.

[LimpingFish]

Interesting set of articles in progress over on RockPaperShotgun from someone working with AI, particularly this one about theft.

The article also links to this Scientific American article on the oroborus eventuality of AI trained on AI.

[ShortWlf]

i did not even see this post before i posted my post inquiring about a toy project i am making in ags, i did some of my art with ai, and others with 3d models, down scaled to pixel art, i did not come out direct, with that being said i did hint at using my own constructed methods, as i have, i do not use ai websites because they are and can be harvested images from copyright or non images, depends on how the ai was trained in other-words, there are a lot of people that just think ai is bad, and it turns into an automatic witch hunt, anyways i think the usage of ai to help design or help get ideas or a trace image, or precursor to develop or make a game is amazing, i just wish people would think of it as a possible tool, rather then automatically label it as bad, anyone remember tracing images in school and using traced images and making into your own art, or using that to help you learn how to draw things?