pouët.net

AI art in compos

category: general [glöplog]
A.I. or supercomputers is not used when chess players play the game in tournaments (unless somebody is trying to cheat of course). But its a great tool to have and to see what the computer comes up with. For example the greatest top chess players use it to analyse their games and maybe before a big routnament they plan their opening or mid moves based on that. So it's a fantastic tool to have learning, exploring the game (learn to make good or strategic moves) and planning. If one understands the point behind the A.I. recommendations that is. Im not saying Stockfish, Komodo etc. is purely an A.I._Engine, but I base my thought process on the general idea. A.I. is not a fixed thing that is. There's not just only one type of neural net etc.

My own thought on A.I. in demoscene compos are the following:
- Don't disallow A.I. entirely, but if you want compos where you obviously know or see that some entry is made entirely by using A.I. and you know there is almost no artistry behind it (code, graphics or whatever), then make have a separate compo for it or ban it entirely.
Lazy people might use A.I. to win a competition, do we really want that? Good people use both human craftmanship and A.I togheter to make amazing demos with soul. There's a line there and I think compo-orgas should have some knowledge behind this philosophical view.

- My personal taste is that A.I. should mostly be allowed in its purest form competitions such as tiny intro compos, because there the artistry lies in the codercraft, one has to code a tiny program using code or data in a way that is different from high-level stuff that use or don't use big libraries that may or may not use big A.I. algorithms. But the tinyintro coder has him/her-self to come up with some clever code and perhaps some amazing new stuff.

I was gonna add more to this list, but I think im done for now, got other things to do.

If it's obvious that some entry was generated with the push of a button, then there is no point in having a compo for humans.

What one should fear is when organizers (compo-organizers for that matter) gets on their lazy side, forget their roots, don't have a clue about A.I. for programming or whatever and combines the A.I. and the humans side into one large soup that nobody likes. Then demoscene is over. Good bye. Coders will be more happy playing minecraft or Quake in their spare-time.

Remember we are humans, not machines.
added on the 2022-08-20 15:46:16 by rudi rudi
I think demoscene is about throwing anything on a given problem that is on offer with regard to ingenuity, technology, creativity, discipline, group effort, etc. For some it's to overcome a machine's or one's own limitations, for some it may be to win a compo even by tricking the audience - whatever motivates. If this includes copyright violation, unscrupulous use of AI, ripping, anim players, that's all great: It's fun, it generates drama, inspires, spurs innovation, creativity and defiance, in a constant struggle to overcome what has been achieved before.
You yourself define the framework within which you want to operate.
The audience develops alongside with the achievements, it may later celebrate what was ahead of its time and hence unrecognized, or identify what they deem cheating, and vote accordingly the next time. There's a problem only if you're taking things too seriously.
added on the 2022-08-20 16:42:20 by bifat bifat
AI image generation is a super interesting development and I believe it's going to allow artists who learn how to use it wisely to be more creative and a lot more productive.

I'm less optimistic about its impact on the demoscene, and particularly on graphics competitions, as the pool of talented artists creating "traditional" artwork (2D, painterly images made with Photoshop or similar, to simplify) has been dwindling for many years already. My concern is that it wouldn't take much for those people to move on if they feel their hard work isn't acknowledged, and if we end up in a situation where the majority of the entries in a graphics competition are AI-generated, there might be a strong feeling of "why bother?".

Oldschool graphics are in a completely different boat, at least for now. We might see even more artists switching to C64 pixelling :)

I'm all for allowing AI-generated images in a separate competition, though. Can't wait to see how the tech can be used in a smart way.
added on the 2022-08-20 18:15:35 by exocet exocet
IMHO AI-art has turn into a very powerful plagiarism-laundering tool, and it will get worse every day: https://promptbase.com/
We’re even starting to see AI-art prompts that explicitly mention ArtStation or DeviantArt… and we’re starting to see AI-art on DeviantArt.
AI-art compos are inevitable, some will probably find an interest in that…

[off-topic]how would we react if demos where created with the help of GitHub Copilot ?[/off-topic]
added on the 2022-08-21 05:54:24 by fra fra
Maybe we should have an "I didn't do it" compo?

- Unreal demo engine
- Additional code by Github Copilot
- 2d art by Midjourney and Dall-E
- Shaders copied from Shadertoy
- Music ripped from an unknowing commercial drum and bass artist
- Faked stamps to be returned back
- Cracked by Skid Row
- Minted as NFT
added on the 2022-08-21 10:16:23 by rloaderro rloaderro
The one that put this together and submitted it yet had the least impact on the final result wins.
added on the 2022-08-21 10:20:19 by rloaderro rloaderro
Quote:
I genuinely hate the short-sighted comparisons between a new technology that expands on human creativity and a new technology that reduces the need for it.

Well, at the time, writing a good 3D engine (read: fast polyfillers) was seen as an achievement in itself. Having a pre-made one (that eventually got really good hardware implementations) negated that; no wonder coders who had spent hundreds of hours polishing their engines got annoyed. But eventually, we learned that even with the same basic building blocks, it was entirely possible to distinguish yourself from the bunch. We realized that indeed, this was increased creative power, not just reducing skills needed. (Also, over time the building blocks become more flexible, but it's not like this changes the fundamental trajectory.)

Here's a thought experiment: If, in 2015, I built something like DALL-E mini myself, from scratch (crawling images, designing the model, training it on some GPUs in my basement—this would be very hard but far from impossible), and used that to enter something in a freestyle image competition, would the demoscene hail me as a great innovator or ban my entry?

In a sense, I do wish it would be enough that people simply marked their entries and then people voted for whatever they found interesting and/or impressive given that information. However, given how aggressive the tone becomes when someone downvotes a Notch prod, I don't really think that's going to fly in practice :-)
added on the 2022-08-21 10:51:10 by Sesse Sesse
Quote:
Maybe we should have an "I didn't do it" compo?

Or an "exploiting loopholes in the compo rules" compo?
added on the 2022-08-21 12:30:07 by absence absence
Quote:
Here's a thought experiment: If, in 2015, I built something like DALL-E mini myself, from scratch (crawling images, designing the model, training it on some GPUs in my basement—this would be very hard but far from impossible), and used that to enter something in a freestyle image competition, would the demoscene hail me as a great innovator or ban my entry?


I think you would be hailed a great technical innovator since creating something like this requires a great deal of technical skill, which you undoubtedly have.

But I don't think the output of your creation would be hailed as worthy of entry into artist competitions, since compos are about demonstrating skill.
added on the 2022-08-21 12:30:20 by farfar farfar
Quote:
IMHO AI-art has turn into a very powerful plagiarism-laundering tool, and it will get worse every day: https://promptbase.com/

Jesus Machine Learning Christ, noone has any shame anymore.
added on the 2022-08-21 13:17:48 by Gargaj Gargaj
Quote:
A Prompt Engineer is a new kind of technician, skilled at crafting the text prompts required for an AI model to produce consistent outputs (e.g. images, text or code) at low costs.


We're saved!
added on the 2022-08-21 13:55:52 by fizzer fizzer
Now I wait for the drama when someone wins an "AI" compo with a picture that turns out to be hand-drawn!
@Pulkomandy: and even traditionally drawn on canvas!
Quote:
Quote:
IMHO AI-art has turn into a very powerful plagiarism-laundering tool, and it will get worse every day: https://promptbase.com/

Jesus Machine Learning Christ, noone has any shame anymore.

But they don't even NFT... what a waste of potential!
added on the 2022-08-21 17:40:40 by D.Fox D.Fox
Technically, AI-generated art is no different than taking couple of source images produced by actual human artists and applying a complex non-linear transformation on top.

So, morality aside, AI-generated art can be considered to be a form of collage. IANAL but from what I was reading online, you can use any art (even copyrighted) to produce collage under "fair-use". What you are not allowed is to sell it without a permission of the original authors, unless... it's transformative (wtf "transformative" really means idk). So, all in all, seems like a grey area.

One of the solution that I can think of is explicitly opting-out from AI/ML training sets usage in your license. It would nice to have option like this in common license templates (like CC), deviant-art etc... So yeah I'm definitely for regulating it. I think many people would still choose to opt-in, but it would be at least their conscious choice.
added on the 2022-08-21 22:50:18 by tomkh tomkh
I was on the Midjourney discord the other day. And while I was surprised by the quality of that stuff.. I was also surprised by the disparity in quality between pictures, or even *users*. Themes, composition, even the style in general. In the end the user HAS some inputs, and some users have more skill (at picking which direction to point the AI at) than others.

I see it a bit like DJ skills, where you don't have to be a good musician, but be good at noticing the good stuff! (And have enough time to sink into that!)

I wouldn't enter a purely generated pic in gfx compo, but it can be super useful to find ideas.
added on the 2022-08-22 05:34:22 by BarZoule BarZoule
TECHNICALLY, AI-GENERATED ART IS NO DIFFERENT THAN TAKING A COUPLE OF SOURCE IMAGES PRODUCED BY ACTUAL HUMAN ARTISTS AND APPLYING A COMPLEX NON-LINEAR TRANSFORMATION ON TOP!

BB Image
added on the 2022-08-24 22:50:50 by gloom gloom
Quote:
Quote:
However, we acknowledge that there is also a craft in creating great prompts for Midjourney/Dall-E

Is there though, really?

No.
added on the 2022-08-24 22:53:05 by gloom gloom
pressing the 'one cup' button on your coffee maker doesn't make you a barista either
you're drunk Bent, go home
added on the 2022-08-25 20:09:30 by farfar farfar
@rez damn, I was too late to post that link :-)
added on the 2022-09-01 10:54:21 by decca decca
Quote:
TECHNICALLY, AI-GENERATED ART IS NO DIFFERENT THAN TAKING A COUPLE OF SOURCE IMAGES PRODUCED BY ACTUAL HUMAN ARTISTS AND APPLYING A COMPLEX NON-LINEAR TRANSFORMATION ON TOP!


I thought the idea was to keep them linear, because of the computational effort increases with nonlinear shit? ;)
added on the 2022-09-01 15:09:35 by NR4 NR4
NR4: right, but also GeLU, SiLU, ReLU, etc...

But of course it's not just "couple of source images".. more like billions, plus actually it's trained to fill between the gaps/noise from all of them sort of "at once", so AFAIU it's very hard to track any source material.

So, yeah, resistance is futile ;P
added on the 2022-09-02 00:29:39 by tomkh tomkh
To me, in the realm of creative work, a competition is about comparing onself and one's abilities across the entire process. From ideation via execution to the result.

Whenever I enter a competition with a piece of creative work, I know how much effort I spent. Resulting in a certain sense of ownership for the result. The more I feel I was able to materialize an idea into a result suiting the quality of my idea bringing my resources and effort to the table, the more I have a sense of having accomplished something. If anything, having persevered through a process requiring a ton of decision-making and committing to them. At best, I learned something new along the way. Or gathered a novel insight. In a way, this is also about personal growth.

What makes me enter a competition in the first place, is the idea of contributing to a context with likeminded people, who share similar values and ideas. About working creatively. To be able to locate myself, my effort, my sikills, processes and routines– in the bigger picture.

While I see potential value in AI generated pieces (and I'm not at the point at which I'm able to call AI results a piece of 'work', albeit it 'art), I think the use of AI compromises the very idea of creative competition in any field. And I certainly don't see why I would want to enter any competiton, that hasn't clearly ruled this out.
added on the 2022-09-02 23:52:18 by rp rp

login