pouët.net

Go to bottom

AI crap in compo entries?

category: general [glöplog]
When a new invention like photography appears, you can say that the photographer is stealing your bread and that no one buys your portraits anymore or you can use your creativity and invent cubism.
added on the 2024-04-02 01:05:23 by ham ham
Quote:
Interior designer doesn’t go stealing the painters income.

But the invention of photography did. So that’s wrong too and must be eliminated?
added on the 2024-04-02 01:12:17 by gaspode gaspode
Quote:

Interior designer doesn’t go stealing the painters income.

It's a different debate.
And yeah, AI bigcorps stealing artists is an important issue.
But I don't think we should throw the baby out with the bathwater.

Quote:

We have to think what is the essence of demoscene and select allowed or prioritized technologies to match it.

During the most active period of the demoscene, the latest technology like graphic and music workstations were not seen in the compos. Using the latest tech hasn't traditionally been the core value of the demoscene.

The Revision PC compo machine doesn't seem lowtech to me :D.

I have the impression than both oldschool and newschool trends have existed and the scene, and that's part of its charm.
It would be sad if the scene missed the AI shift and that the oldschool part became more and more prevailing.
added on the 2024-04-02 01:15:52 by wullon wullon
Quote:
When a new invention like photography appears, you can say that the photographer is stealing your bread and that no one buys your portraits anymore or you can use your creativity and invent cubism.

fuck yeah!!
added on the 2024-04-02 01:17:31 by wullon wullon
paraphrased: "why should I be bothered to vote on a prod that nobody bothered to make?"
added on the 2024-04-02 01:29:38 by vurpo vurpo
Quote:
Quote:
I think it's time that we get some seminars offered by scene artists who have been using AI in creative processes.


As an introduction to "AI" / machine learning (ML), I found @blackle's ShaderToy entry quite interesting. She also recorded a YT video (see link in the entry description) which explains what's going on / how it's done.
The shader mainly consists of the network's tensor matrices / activation functions (sin) and is generated by a script.
The input is the screen position and the output is the (approximated) signed distance to the 3D object the network was trained for.
It's a good example because it is rather simple (on a side note: the original, uncompressed object file is ~22mb uncompressed, the (naively, just for testing) 7z -mx9 compressed, model-specific GLSL code (scene() function) is ~2.5kb)

TL;DR: train your own networks


this is a really good example of the difficulty in even talking about this stuff (maybe by design!!). a novel algorithm on a public domain test object crammed into a 4k shader? great stuff, very cool, good use of technology. that's worlds apart from what midjourney &co is doing. I think we need to be very clear about what's being talked about here. The major difference ofc being the corpus used but also, what you're doing with it
added on the 2024-04-02 01:38:01 by hornpipe2 hornpipe2
it's amazing how every single time this issue is brought up, the discussion is instantly muddied and diluted with hundreds of unrelated/tangential comments like "but what about..." "well you could also imagine that..." "but what if someone did this instead" "you could even claim that..." and any actual thread or point to the discussion is immediately lost to the noise.

It's a very effective way to shut down anyone trying to say anything against "AI", just make everyone give up on trying to say things in the first place!
added on the 2024-04-02 01:47:34 by vurpo vurpo
@hornpipe2: when ppl read "AI" they hear "artificial intelligence that will make you obsolete" when it's actually just "algorithmic interpolation", as in polynomial approximations of function segments, just a tad more advanced.

the copyright issue of the sourced training-set data is a whole other topic, though. I recently wrote a lengthy rant about this and the bottom line of it was: It is not a free-for-all ! list your sources, license your content, then we're all good.
added on the 2024-04-02 02:15:43 by bsp bsp
@vurpo: You can say things against AI, because it’s an laaaarge topic. This discussion here is about AI in art. But there are other fields where imho AI is a much more terrifying topic, like voice simulation, deepfakes, letting AI driving cars, trains, drones, using AI to suggest targets for war. Just because in this specific area of art-synthesis I don’t have so much problems with AI doesn‘t mean I don’t see dangers in other areas.
added on the 2024-04-02 03:27:48 by gaspode gaspode
Here's my view. I see (for example) decoupage craft having a lot more to do with visual arts than prompt jockeying. Prompt jockeying comes off as more akin to SEO or gaming the search engines. Sure, visual artist could use AI generators too. But why? He can use whatever he/she likes, p*ss and sh*t too, as already mentioned. Why wrestle with verbally gaming the machine instead of doodling with a pencil, really practicing art/craft? Because tech bros said so? Who will feed the machine with new stuff when all we create is going to be made by that same machine? Don't you see the extreme averaging this brings? The noise, the entropy that will inevitably overtake this closed loop? Remember, the intended use of AI is not to help the artist, it's to eliminate the artist.
Anyway, I think this thread ought to narrow down. Go back to the start. Which is "AI crap in compo entries?" With a question mark. I guess the original prompt was "How do you feel / what do you think about people entering AI made stuff into compos against 'manually' made stuff?"
added on the 2024-04-02 08:54:33 by 4gentE 4gentE
Well, a portion of the demoscene graphics branch has always been a "Mom, look, without needing inspiration, I made an image similar to hundreds of others found on the internetz by pixelating over someone else's work". People have won graphics compos in the past and keep doing it nowadays going that way. So, I personally don't see much difference between that and a "Mom, look, without needing inspiration, I made an image similar to hundreds of others found on the internet by pixelating over an image generated by AI" situation. People (let's say) with little or no inspiration will keep on doing what they do, enter compos with boring themes, girl faces and cutesy animals that most people understand and like. Occasionally throw in an Amiga ball to render the otherwise random image 'demoscene-ish'. =D

Also, Tesla and Horse in same race is pretty much happening in the demoscene. Just look at oldschool categories!

Also, decorators are not artists.

Also, this: "AI art will be really interesting when AI is creating it for itself and doesn’t need the humans anymore. :o)"
Nobody on the scene is getting paid, so that discussion is pointless in this context.

In general, AI generated images in demos suck, because the images themselves usually suck. Especially on platforms like the Amiga, where crappy downsampling is inevitably added to the generic, instantly recognizable "style" of images that usually don't stand up to even the slightest of scrutiny.

Cubism doesn't have to be invented. Real pixel art techniques already exist and produce infinitely better results. Real individuals with actual, recognizable personal styles already exist. Inspired artists with the capacity to reason about platform limitations in interesting and creative ways already exist.

There are artists producing entire suites of thematic images, playing with variations on perspective, colour or a particular motif. There are artists who make me do a double take when seeing a 32 colour image in 320x256, because it truly looks like more colours and pixels than there are. There are artists working wonders with extremely limited resources, fitting things into dual playfield limitations or cramming highly detailed marvels into C64 hires. There are skillfully drawn and composed images that, when you start really looking at them, reveal little details like star- and heart shapes in the dithering, making your smile widen or jaw drop. I pray they're not dissuaded by the apparent indifference towards all of this, because theirs are pictures I love to look at.

But hey, I guess the scene was never about pushing your technical and artistical skills. Just fill it with more of the safe blandness people recognize from their infinite scrolls of trite nonsense designed to make them watch yet another ad, and they feel right at home. Attention to detail is for suckers.
added on the 2024-04-02 10:32:59 by grip grip
please do try to keep a clear distinction between "AI" as a technology (DLSS! depth extraction from images! simulation upresing! nerf generation! lots of potentially interesting use cases that may or may not affect the demoscene) and "using generative AI tools" (the bit where a few massive corporations combed the internet, grabbed all of human creative output and then sold it back to us).
added on the 2024-04-02 10:33:16 by smash smash
It helps to take the emotional "but commercial photo AIs have been trained on 'stolen' art" part out of it. Will AI put lots of creative people out of their existing jobs? Likely. Might AI destroy trust in news? Sure. Might AI kill us all? Possibly.

But it is THERE. Ignoring that it's there and that it's used just would make the scene be stuck in time, with innovation happening elsewhere. We can't forbid it, and we can not prevent it being used inside the side, so we will have to deal with it.

As we have done in the past, we will have to adapt, adopt and improve. We survived Data Becker's Demomaker, and we have survived web demos, and we have survived point&click demo engines.

Also, AI is not just about generating plagiarism.

What if I am a good musician, but very bad at mastering my track, and I don't have anyone around that could help me with the mastering? What's wrong with using a mastering AI, as long as I disclose having used this tool in the streaming music competition and having been unable to use it in the tracked music competition?

Or when it comes to visual art - what if I have a great story to tell, or a cool effect to make, but don't have access to a 3D artist? Is it really that bad to use an AI to help me generate meshes for use in the demo, as long as I disclose it? We had lots of demos with the 3ds duck, where the point was not to claim "I have created the 3ds duck".

We might one day end up in a situation where it will take me 5 minutes to create a fake ASD demo created and you no longer being able to judge if this really is ASD or not. We aren't there yet. And even if we were, the scene is about freshness when it comes to styles, and people probably would not believe me to be the original ASD, anyway.

To repeat myself: Personally, I just want to know what tools, including AI, have been used to create a production. I want to know how much is pre-existing stuff based on other peoples work (which, again, includes the demo engine used, because a coder might have spent years to create their own, while other groups may have used a point&click tool), and how much is original content. Correct crediting is all I need to be happy.
added on the 2024-04-02 10:44:07 by scamp scamp
We didn't mention AI tools because they were not part of the artistic process, and why would we disclose it anyway? We were condidering to mention Gimp, AKlang, Wave Manuel, vasm, Kate, KDE, Richard Stallman and the Dalai Lama (in roughly that order), but decided against ... because not important
added on the 2024-04-02 10:50:40 by bifat bifat
We designed and trained our own neural net for the Skilander to convert whatever 1bit characters to duke nukem 3d-ish multicolored tiles. It works by convolving the 1bit alpha masks multiple times with learned (optimized) kernels. Read all about the motivations and details from Dukenukemifying.the.chargen.ppt!
I'm thinking of adding a "Generative AI used in this prod" option to PartyMan which the person submitting the entry will check or uncheck, and to show that status in the entry slide and voting listings. Seems only fair to let voters know.

Lying about that is obviously a breach of rules and will lead to disqualification.
added on the 2024-04-02 11:18:28 by T-101 T-101
Op een mooie zomerdag was Dikkie Dik lekker aan het spelen in de tuin van de familie Jansen. Hij rende vrolijk rond, klom in bomen en speelde met zijn favoriete speelgoed. Plotseling voelde Dikkie Dik een dringende aandrang. Hij moest heel nodig naar het toilet!

Dikkie Dik begon te miauwen en rond te rennen, op zoek naar een geschikte plek om zijn behoefte te doen. Maar de deur van het huis was dicht en hij kon niet naar binnen. Paniekerig keek hij om zich heen en zag toen het tuinhuisje aan de andere kant van de tuin.

Met alle kracht die hij had, rende Dikkie Dik naar het tuinhuisje. Hij duwde de deur open en vond eindelijk het toilet. Met een zucht van opluchting sprong hij op de wc-bril en deed zijn ding.

Na zijn dringende behoefte te hebben gedaan, voelde Dikkie Dik zich opgelucht en tevreden. Hij waste zijn pootjes in de wastafel en liep toen trots terug naar de tuin, waar hij verder kon spelen en genieten van de rest van de dag.

En zo leerde Dikkie Dik dat het belangrijk is om op tijd naar het toilet te gaan, zelfs als je midden in een leuk spel zit. Vanaf die dag zorgde hij ervoor dat hij altijd wist waar het dichtstbijzijnde toilet was, voor het geval dat dringende gevoel weer zou opkomen.
added on the 2024-04-02 11:20:13 by havoc havoc
@havoc. Is Dikkie Dik the father of all the happy cats we've seen in demoscene products over the years?
resbeng: Unfortunately I cannot confirm that Dikkie Dik is the father of all the cheerful cats that can be seen in demoscene productions. It is likely that different artists and creators have their own sources of inspiration for creating characters, including cheerful cats. Dikkie Dik is a well-known character from children's books and may certainly have been a source of inspiration for some artists, but it is difficult to say whether he is the father of all the cheerful cats in demoscene productions.
added on the 2024-04-02 11:32:17 by havoc havoc
Thanks for that. I really was wondering there whether Dikkie Dik might actually be THE cat behind the very foundations of the demoscene.
Quote:
We didn't mention AI tools because they were not part of the artistic process, and why would we disclose it anyway? We were condidering to mention Gimp, AKlang, Wave Manuel, vasm, Kate, KDE, Richard Stallman and the Dalai Lama (in roughly that order), but decided against ... because not important


well, you made us look at your dumb shit for ages, if you had put some more effort in the images people might have given the tech props. Shame really.
added on the 2024-04-02 12:34:56 by okkie okkie
I started this project for the art, not for tech. We had some tech lying around from older projects, but it proved insufficient during the process, so we were glad some tech titans from the C64 scene helped us out.
added on the 2024-04-02 12:49:50 by bifat bifat
OK, no money involved. So, regarding the original question: a lot of people seem to be saying: Look, if the audience of a compo likes some AI generated crap more than a manually pixeled image, who are we to judge? Did I get that right? I understand the point. But it makes me wonder: How do you suppose pixelers will cope with that? You know, the ones who's work this despicable machine had been trained on? How will you keep them invested? Is it worthwhile to simply dump them just so that a bunch of us can feel like artists?
added on the 2024-04-02 12:51:26 by 4gentE 4gentE

login

Go to top