pouët.net

AI art in compos

category: general [glöplog]
Quote:
Example -- an animation director's job is not to animate, but quite literally "ask somebody to paint something based on their description", and definitely is an artists.


imo its closer to "the client" than a higher creative role like a director. because the relationship is more "show me something based on my rough brief, i dont like it, show me something else until i do, make it cooler, make more real, copy this artist now" which is familiar to many designers as a client relationship :)

the director in your example comes in with a creative vision, they know what they want to see and guide the people below them to create it. they can often visualise before starting what they want to get and work until they get it. maybe some people do use ai like that?
added on the 2022-12-24 10:54:14 by smash smash
Quote:
I would say the decisions you make •during the creation• of something is what makes you creative. With AI, those decisions are not made by you - they’re made FOR you. In effect, you’re just a client.


a neural network is just an algorithm that is not aware of itself and thus doesn't make any decisions. it is even a deterministic algorithm that gives out exactly same output if the input is same. there is no any kind of decision making there.
added on the 2022-12-24 11:49:29 by nosfe nosfe
Not in a human sense where you weigh consequences, but in order to reach any kind of result, decisions need to be made - so yeah, AI’s do make decisions. Self driving cars make decisions. ChatGPT makes decisions when you ask it to come up with a short story.
added on the 2022-12-24 12:48:17 by farfar farfar
Pushing the button on your coffee machine does not make you a barista, also not the decision to push the button.
added on the 2022-12-24 13:10:06 by gopher gopher
using other people's analogies in jest is as bad as pretending to be a skilled driver in a self-driving car
Hey it's OK if you're not very skilled in the art of getting a clue. Not everybody is! It's perfectly normal.
added on the 2022-12-24 18:06:36 by yzi yzi
Not in a human sense where you weigh consequences, but in order to reach any kind of result, sdecisions need to be made - so yeah, AI’s do make decisions.

decision between what? stable diffusion and other image creation tools are based on deterministic algorithms, you get exactly same output if the input is the same. where do you see decision making there?

Quote:

Self driving cars make decisions. ChatGPT makes decisions when you ask it to come up with a short story.


a tesla car going rampage and killing someone might be close to being an art performance, but I would not give it credit for that just yet.



people having a 19th century understanding of the definition of art is way too common so I'm not surprised to find that here, but I would expect people to have a better understanding of algorithms.


Quote:
Pushing the button on your coffee machine does not make you a barista, also not the decision to push the button.


i've understood being a barista means you are an unemployed artist if you are living in usa so at least it's connected to creativity!
added on the 2022-12-24 18:48:15 by nosfe nosfe
Quote:
Pushing the button on your coffee machine does not make you a barista, also not the decision to push the button.

Well, do you aldo have to roast the coffee by yourself to be a true barista? Where do you draw the line?
added on the 2022-12-24 19:22:48 by v3nom v3nom
Sounds more like we don’t have the same definition of what decision is
added on the 2022-12-24 19:34:56 by farfar farfar
Would it be possible to develop skill and taste to use AI image generators so that it becomes an art? "The art of making images with AI." Not possible?
added on the 2022-12-24 20:34:54 by yzi yzi
oh, it's art.. just not artisan
Quote:
Sounds more like we don’t have the same definition of what decision is


i would go for dictionary definition
"a choice that you make about something after thinking about several possibilities"

a deterministic algorithm, like stable diffusion, does not make a choice between several possibilities.
added on the 2022-12-24 21:24:03 by nosfe nosfe
I’m sure there’s an interesting discussion about the difference between a deterministic algorithm reaching some result given an unchanging set of parameters, and a human agent reaching some result given an unchanging set of parameters.

.. but I was thinking more of self driving cars - those don’t make choices?
added on the 2022-12-24 23:34:26 by farfar farfar
they don't. they make equations.
I just made it through this thread half-way.

I'm not the philosophic type and when browsing through a lot of AI stuff, there was just one thing coming to my mind related to the demoscene, no matter if this is a 19th century art opinion or not:

Would I want one of the parts of the following sentence come true:

"Let's just create demoscene art by A.I. and run it in compos organized by A.I. and let A.I. vote for it and be a spectator."

Nope.
added on the 2022-12-25 08:21:15 by Raven^NCE Raven^NCE
i think there should be a special compo section for AI Graphics coz it´s already there and we can´t close our eyes and ignore it.

The Compo could, as someone said, handled like the photo compo for the "lower efforts" section. I agree that it´s zero to compare with a demo, intro, a pixeled graphics or something else. It´s one of the easier tasks if not the easiest to create "arts"

is it art.... i think YES. Should it have a high value like painting or pixeling etc. NO

So would it hurt to include this kind of art in a very own competition.... i say no but you need to decide on your own.
Is making screenshots from 3d games an art? You make a choice on camera location after all. You should be able to print it, sign, frame it and sell to art snobs ;P
added on the 2022-12-25 16:31:41 by tomkh tomkh
Whether the output is art or not is a complete diversion, and only two things should be in front of you when it comes to image synthesizers (which I will continue to refer them as, since AI is too broad of a term):

1. People who make image synthesizers are not your friends

It's a good idea to start with the following thought: We've reached a point in history where technology should not only based on its merits, but its intentions and context - not just what was created but WHY it was created? What was the vision of the people who created, and does it align with yours? In the last decade or two, we've been inundated with technology (largely in forms of social media) that offered a sheen of utility, but that ultimately revealed itself to be a tip of a very large, and very uncaring iceberg.

With that in mind, it doesn't take a whole lot of looking around to reveal their intentions and motivations, and they make it very clear: none of us factor into any of it, and once they start to move fast and break things, they will not stop for you or me - after all, they already started with questionable legal grounds. (Which is why, by the way, the capitalist apologia in this thread is hilarious, especially when it comes from angles you wouldn't expect - hi Nosfe!) Let's face it: image synthesizers are made by companies stuffed to the brim with VC money, and they're functionally no different from a paperclip maximizer eating everything in its way. Sure, they let you play around with DALL-E and ChatGPT, but you're just feeding the machine. (Also literally, as your inputs get integrated into the larger dataset.)

We don't have to be part of this. I was heartened the recent near-unanimous rejection of the crypto/NFT bubble within the scene context, which ultimately turned out to be the right stance in history and we never got tainted by the utter embarrassment it turned into. I hope we do the same here - the writing is on the wall - but while NFTs always smelled funny and there was no technical utility, it worries me that the shiny polish of the image synthesizer output taints a larger crowd this time.

2. All of this is antithetical to the core principle of the scene of seeing what others can do

There's something to be said about WHY so many people are excited about it in (around) the scene, and I think it's time for some introspection on the whole "it all started as a popularity contest" angle, because while it is true that we had crackers putting their names in front of games, we had charts, and the groups in the scene who are good at things still get a degree of (limited) popularity out of it, but the understanding was that all of these things were a metric of the craftsmanship, the artistry of the people you know, and productions were never meant to be a means to an end where the end goal was to be number one. Image synthesizers offer very little insight into either the artistic process of whoever entered the prompt, and is entirely devoid of the actual craftsmanship of someone developing an eye for detail, perspective, lighting, shading, all that sort of stuff that takes years to learn: one my biggest joys in the scene has always been seeing a competition and watching people enter their first close-to-zero janky thing, but then hone their skills over time, learn more tools, open their minds to various influences, create their own style - just watching the process of GETTING BETTER and BECOMING SOMEONE gives me the joy of seeing what it's about.

Here's the thing: We already dropped a similar ball with Unity / Unreal Engine when it comes to coding - I do believe it was in good faith, to see if they're viable springboards for people who didn't have the ability to put a demo together from the ground up, but we unwittingly joined the games industry in what I can best describe as a downward spiral in talent outside those engines, and the dwindling number of custom engines, with artists pressuring teams to very quickly adapt to the UE level, lest they'd just use that instead. The trajectory isn't good, and image synthesizers are simply the same trend from another perspective - general purpose engines made coders obsolete, image (and eventually audio and topology) synthesizers may do the same to artists - in what I imagine to be the worst case scenario, we'd end up with coders making demos with synthesized art, and artists making demos with commercial engines, instead of the two working together, which would be tragicomical. (By the way, I would actually totally appreciate if someone would develop their own image synthesizer together with an artist using their dataset as an input, and would produce dope pictures with them - which would really be the scene thing to do - but there hasn't even been a suggestion for this, which gives you an idea where we are on the above spectrum.)

(Tangent: I also don't see the point of discussing whether putting a prompt in is art or not - because trust me, you won't get to do it for too long: If your contribution is being the "idea guy", only coming up with a text string - even if it's heavily parametrized - best believe you don't matter much in the process, and the same principle that can generate a high definition image sure as shit can produce the string needed for generating it, especially for an incredibly transparent community like the scene: How much work do you think it would be to create a bot that checks out Demozoo for recent graphics entries, maybe reads Pouet for some recent trending topics, figures out what's popular, formulates it as a sentence, feeds it to an image synthesizer, and then emails it to a party organizer? See - no need for your "prompt artistry", and the shoe will very soon be on the other foot.)

Maybe I'm an idealist softy who cares too much about the actual people behind the work, but image synthesizers (and the people who use them) offer no insight to the history of an artist, the skill of a coder, I can't be friends with a pile of Python code, and thus (in my eyes) they are far from the spiritual core of the scene.


So where does this leave us?

I don't like the idea of treating it to its own compo (though it's been done before, and that's fine as a toe dipped into water, but I'm wary of it becoming a trend), because the argument that the only two options is to either ignore it or accommodate it feels like a false choice to me: Just because it's currently rampaging through the world doesn't mean you have to submit to it, and moral rejection is always an option. And I get it, since they can generate steps now, it's going to be near impossible to keep a graphics competition AI-free, but it's the principle that matters, because in the end how would you rather describe the scene? "Hi, we're the demoscene, we type single sentences into commercial proprietary code" or "Hi, we're the demoscene, we are interested in what human talent can produce"? What do you want the scene to communicate? Because if it's "everything goes", then you're just a generic purpose internet forum, and not a particularly popular one at that either.

Let's not let a quick dopamine high take our eye off the ball.
added on the 2022-12-25 17:32:24 by Gargaj Gargaj
If they are having fun doing it, keep at it.
added on the 2022-12-25 18:35:00 by thec thec
I guess posting that amazing wall of fallacies gives its author a good dose of dopamine!

Thank goodness right now there are young people learning to use these new technologies in a creative way and they won't care about the catastrophic opinion of angry old farts. :D

Neither the world nor the demoscene are going to end in the near future, even though many John Connors come out saying otherwise.

@Gargaj: Don't be so pessimistic. Try Stable Diffusion locally, you can do it even in a laptop without internet connection just using CPU and RAM. Everything is open source (you are right about systems like DALL-E that are closed but not about Stable Diffusion) and you are being unfair to your frenemies talking like that about this technology in general. Concerning the compos: As I said in a previous post, soon everybody will use latent diffusion methods as another tool during their workflow cause every major graphic tool will have at least a plugin for using some latent diffusion technique (with the possibility of train your own neural models). But you can forbid, if you want, for example the use of the circle tool in compos and force each participant to demonstrate that they can draw a circle pixel by pixel by themselves. Because, obviously, you are not an artist if you can't... right? Good luck with that.
added on the 2022-12-25 19:17:36 by ham ham
Quote:
Everything is open source (you are right about systems like DALL-E that are closed but not about Stable Diffusion)

Oh sure, I'm the one posting fallacies.

I never said they're not open-source. I said they're for profit.
added on the 2022-12-25 20:44:38 by Gargaj Gargaj
I think NFTs we could all smell its rotten origins. In this case however AI art does have lots of uses; say for a making a logo for your little caffe, hair dressing saloon or whatever business without having to hire a designer. Even if in the first incarnations of the AI tech the output is worse than what a professional illustrator delivers, people will flock to the cheaper solution and wait for the tech to catch up. The fashion industry went through this, text translation too, and many other. Art/Design is next.

My problem I have with AI art is not that, but that the tech comes yet again from a crowd that doesn't have respect for artists (or anybody really). The fact that they can one-sidedly decide to scrape the internet and collect artists, artisans' and creators' work to train their models, commercial or not, without asking let alone paying them, enrages me. They think creative's labor is free to use, just because it is free to access. Or if they think it's actually not free, they bully and still take it anyway.

When you bring these criticisms to them, and I have done so openly, they explain that what they do is not illegal, which is the answer you'd expect from somebody confessing their moral wrongdoing, confidently.

I think most people our age would agree artists should be respected and allowed to opt-in into licensing their work, or donating for AI training purposes if they want so, but never forced to accepting it's been taken from them for free. The fact that even opting out is difficult or just impossible is sad.

I also think that from these company's perspective, there are strong incentives for not behaving ethically, because the perceived competition among themselves means whoever takes the ethical shortcut collects a millions of times larger dataset and is at a massive technical advantage over those doing the right thing (ie, asking and paying for the content).

If you criticize this short term greedy behavior, they'll surely claim that in fact they think mostly in the long term and do see AI art as eventually a net positive for society (which I agree with), and that therefore short term casualties, sad as they are, don't matter much in the big picture of things (which I disagree with).
added on the 2022-12-25 21:06:07 by iq iq
Wrt the original topic: we, the humans, don't allow AIs on chess compos so why would we allow them on demo compos? It’s the exact same thing right: computer gets good at something only humans could previously do, humans boot the computers from their compos and hunt down the cheaters, and everybody carries on merrily.

I dont see how the “VC funded = evil” or “trained on copyrighted” arguments factor into that. They’re fun discussions to have but if all AI was made for free by unpaid hobbyists and only trained on deviantart pictures with a “ok for AI” checkbox checked, nothing would be meaningfully different wrt its use in demo compos.
added on the 2022-12-25 21:50:25 by skrebbel skrebbel
What Inigio says about "legality" of collecting source material is of course correct. Unfortunately, creators, scientists and even coders gave away the fruits of their work pretty much for free to "wrong hands" long time ago. And even you guys are perpetuating it, e.g. ShaderToy is a double-edge sword. Now, it's trivial to scrape all of the shaders code and train the language model for "AI" shader generator tool ;P

But for me personally, there is a bigger problem. I'm quite disappointed by capabilities of those so called AI tools. Stable diffusion, GPT, etc... are not capable of producing new content. They are just good at mimicking and creating variations of existing content. It's unfortunately already good enough for many use-cases. If you think about it, most people are also good at mimicking and only a few are able to create truly original content. But please keep calm, big tech don't have any clue how to create true AI. It requires mad genius, Nikola Tesla's level, to solve, who will be hopefully smart enough not to share the solution with idiots. Unfortunately, it's gonna be hard...Edison's of today's world are already in full alert :P
added on the 2022-12-26 16:00:16 by tomkh tomkh

login