pouët.net

Go to bottom

AI crap in compo entries?

category: general [glöplog]
Pretty sure the plot and text is 100% human-made. =)
added on the 2024-04-09 22:43:33 by Krill Krill
Quote:
I've seen the Demozoo screenshots of TM2 and its development linked above, but I'm still nowhere near understanding HOW and WHERE AI played a role in this! All I see is comic book panels that look human-created, so apart from the strange plot in the speech bubbles, where else did AI have a role?

Everything was done by Greippi, including text and plot. At some point the panels were colourized in Gimp with material obtained via AI.
added on the 2024-04-10 00:25:41 by bifat bifat
So most of the work was done by human talent (Greippi) with a little help from AI with Gimp colouring in the panels, and people are objecting to that latter part, in, I suppose, speeding up the production. Was there a deadline to meet?
added on the 2024-04-10 00:47:52 by Foebane72 Foebane72
Of course we have set ourselves a deadline, even a tight one. The use of AI was not a shortcut. Everything here was planned and fully on purpose. It was the same process as for TM1, just refined and enriched with more tech and music.
You can call it an experiment (and it's still running. :-)
added on the 2024-04-10 01:05:19 by bifat bifat
Quote:
So most of the work was done by human talent (Greippi) with a little help from AI with Gimp colouring in the panels
Afaics, AI provided the visual "assets", which is far less straightforward than many people seem to believe. If you read the "making-of" document and take a look at the accompanying "raw" AI output, the latter had to be cut apart and collaged in a process that turned out to be no faster than the traditional way. There's only so -> <- "much" context you can transport in a prompt, or such goes my head canon. :)
added on the 2024-04-10 01:40:13 by Krill Krill
Quote:
So most of the work was done by human talent (Greippi) with a little help from AI with Gimp colouring in the panels, and people are objecting to that latter part, in, I suppose, speeding up the production. Was there a deadline to meet?


"A little"? :) If you are extremely good at drawing and painting this is still weeks of work. If you are bad a drawing this is beyond your reach to even get close to this level.
I wouldn't call this "a little" help with the AI making this.
added on the 2024-04-10 08:30:09 by The_Sarge The_Sarge
Quote:
True, but it's a theoretical possibility. Here's a wilder one though. What if, in the future, it becomes possible to generate pictures with AI with sufficient speed to turn them into animation instantly? That would be an ai being able to churn out at least 25-30 images every second. Could an AI like this be considered a demo?


This is the way...

And we are not talking about the future.

https://github.com/IDKiro/sdxs

https://old.reddit.com/r/StableDiffusion/comments/189onqs/today_i_hit_77_images_per_second_at_512x512_with/

The current record for image generation with SDXS in 512x512 is something like >400fps on a RTX4090.

There is some fascinating technology in GenAI and its well worth looking into this also for demos.

Maybe some of thouse could become the next raymarcher?
added on the 2024-04-10 08:33:45 by Azure Azure
Quote:
I wouldn't call this "a little" help with the AI making this.


Me neither. lol.
added on the 2024-04-10 09:26:56 by neoman neoman
Quote:
Please explain, cos I'm lost.
Have you ever used an AI image generator? If not, I suggest you play around with one for a bit. No matter what one's position might be on using AI, it's very apparent to me, even at a quick glance, that these graphics were AI generated. Often - though not always - the images get a certain quality, a specific look, that's easy to spot. And that's even after the fact that there's been quite some work in curating and editing in the production discussed here.

Quote:
There's only so -> <- "much" context you can transport in a prompt, or such goes my head canon. :)
Yes. Prompting is the reverse of the idiom "A picture is worth a thousand words". When creating an image using traditional tools, the artist has complete control over everything. AI prompting is more of a suggestion, really. You can tweak the prompt to a certain degree of course, but as soon as you want control over specific details, things get frustrating. And if you start pushing too hard in one direction, the other bits fall apart fast.

It's fun to play around with, but it's a very limited tool when it comes to perfect realization of a fully formed artistic vision, even when it comes to a single still image.
added on the 2024-04-10 10:34:29 by grip grip
IMHO the outcry is disproportional to the actual risk..AI bubble seems to be breaking already. Room for improvement is slowly diminishing. I bet it's gonna join NFT or Metaverse fate soon and then it will be suddenly cool to use AI in demos ;-P
added on the 2024-04-10 13:08:09 by tomkh tomkh
I hope some people (with actual Amiga knowledge) will actually watch TM2 itself at some future point and indulge in the fact that generative AI was used (while the prod still wasn't trivial or quick to produce at all)... before taking a dump on the prod's entry page. =)
added on the 2024-04-10 13:15:01 by Krill Krill
Yeah, the prod definitely did not get enough attention yet.
added on the 2024-04-10 20:04:32 by v3nom v3nom
Quote:
Yeah, the prod definitely did not get enough attention yet.
Entirely beside the point.
added on the 2024-04-10 21:17:17 by Krill Krill
Quote:
with actual Amiga knowledge

What difference does that make?
added on the 2024-04-10 21:54:07 by absence absence
Quote:
What difference does that make?
That was meant to refer to some specific people who did take a dump on the entry page.
(Otherwise, well, might make a difference wrt assessing the mere technical merits.)
added on the 2024-04-10 22:28:05 by Krill Krill
665
added on the 2024-04-10 23:11:39 by aqu aqu
666
added on the 2024-04-10 23:11:45 by aqu aqu
So reading through this again and skipping over some exaggerations, insinuations and noise from the dutch troll brigade, the gist here is:
Use of AI even in this strictly non-profit environment changes the way someone perceives art, this reacts upon the yes-profit environment, where in the long run it destroys the jobs of illustrators.
In addition to that, it infringes on copyrights today here and now, or in other words: Artists and illustrators do not realize their due profits.
Would you agree, or summarize it differently?
Who was hurt?

And then of course, what will Hans Bliderman aka Toxic Modulo do about it?
added on the 2024-04-11 00:03:07 by bifat bifat
Using AI art for "aux assets" or processing doesn't really affect the artistic value of production imo. Couldn't care less if textures are AI generated instead of being first google image search hit, character animation is AI generated instead of downloaded from mixamo, overlay graphics use AI upscaler or demo soundtrack voice lines are AI generated.

We constantly borrow and modify stuff anyway and it's expected. Only when your show-off moments are AI generated and it's not clearly stated or obvious it becomes questionable.
added on the 2024-04-11 08:41:04 by sauli sauli
i've used AI stuff in my "graphic design" collages, but at that point I use it as a paintbrush. Oldschool demos use it as an actual scene which is ridiculous & lamer.
added on the 2024-04-11 11:39:25 by wrighter wrighter
Please elaborate on the "AI paintbrush" thing with collages.

Each of the panels is a collage composited from various AI-generated pictures.

"Oldschool demos" plural? Do we have a trend (3+ comparable instances) already?

And downthumbs go there. =)
added on the 2024-04-11 12:30:45 by Krill Krill
If you need a picture of Titanic sinking, are you going to staff a full boat and run it into an iceberg or are you going to use Google image search or AI?
added on the 2024-04-11 17:00:18 by rloaderro rloaderro
depends on who's paying
added on the 2024-04-11 17:20:55 by havoc havoc
well this has gone way on in the days since I posted it.

a couple things stand out:
* sorry for just picking on TM2 in my original post - as it was the most "obvious" example of blatant AI image gen on quick watch. As others have pointed out, Amiga winner "The Deep Meet" (https://www.pouet.net/prod.php?which=96598) allegedly contains unmentioned AI assets (?), and ASD's second-place PC demo also uses it (but that's mentioned in the prod readme / credits). Plus prods from other parties going back a year or three.

So it's not just hand-wavy as "who cares, TM2 took 8th place after all". That's true but AI in prods also took first, and second, and who knows what other places. Adjust your arguments accordingly.

* I don't like the statement of "if it's so internally fulfilling people will keep doing it!" Yeah, I do find personal value in making art, but I ALSO find a lot of value in making art other people want to see, and getting positive responses from other people! They're COMPETITIONS, after all. Trying to discount extrinsic popularity as some less-than-pure motivation is just silly. If I was doing something people formerly appreciated, and they stopped because a machine took my place, I'd not stick around... and whether I go paint in my room alone or I just give up entirely isn't important, really, to the conversation at hand. Accept that people have different motivations and proceed from there.

* I guess as a minimum I would like to see AI usage mentioned a "making-of" tools section, in the same way you might say "public domain space images" or "Unreal engine" or "Renoise". Everyone is agreeing that the audience will never suffer from more context provided.

Obviously, some people may lie or not state it. In that case, disqualify them if caught, just like how prods have been pulled off Pouet for including a commercial audio track (http://www.pouet.net/prod.php?which=30284). A partial, socially-enforced solution is good enough when a technical one isn't possible.
added on the 2024-04-11 21:21:07 by hornpipe2 hornpipe2
Quote:
If you need a picture of Titanic sinking, are you going to staff a full boat and run it into an iceberg or are you going to use Google image search or AI?


Ridiculous. But exactly as described in the quote, in 3 years, someone will think such a thing beyond human reach, while all major newspapers of 1912 will have paid for and featured an illustration of a boat and an iceberg, drawn by a human. Many history and engineering books and magazines will since have paid for and featured new pictures, drawn by other humans, through the decades.

Quote:
Trying to push out generative AI by putting more pressure on people is probably a very bad idea.


Of course it adds pressure, because the AI obviously doesn't know what anything IS, so how would it know how to draw anything? It's fed a bunch of already drawn pictures and matches prompts. AI itself can be used for good or bad. It can be used for stealing art. That's bad.

Quote:
Do we need a new word for productions not containing any AI stuff?


Just lazy prod for one containing AI images is fine. I would feel the same about converted pictures (found online or otherwise). To a lesser extent about copy (e.g. Shaft No.5 in legendary OCS Amiga demo). There are rules in gfx compos to find out who are artists and not. The same rules could apply to gfx in demos. I think the artists who actually draw pictures for gfx compos should have the greatest say on the topic of AI "generated" images.

Stuff, yes. The criticism against AI when it comes to things that look like art is really the same as it has always been: If it looks too much like a photography, you question it because a photography is not a painting. If a painting looks to much like another, you question it because it's close to a copy or at the very least derivative.

AI stuff may well force a definition of what art is. To me, it's a conscious expression by the artist, drawn by the artist. And this, combined with: "Ideas are cheap". I have better ideas than any prompter (or a list of prompts generated by AI from popular searches for prompts around the web). Ideas don't really matter, until the one who came up with them paints the painting to match. And you know as well as I that some ideas require many tries. This is what we look up to. A great idea, implemented. It's what makes art,art.

The pictures you see are generated from many ideas from many artists, some of which may have painted the pictures themselves. And there are photos of kids, as you can see. There's no credit for any source in the amalgamate of all the searches. This is the reason for the artist's signature. (There are cheaters here too, current and past masters hiring others and signing their own name because they paid something called money. Part of the definition of art, is that this equals not art. The artwork is immediately dismissed.)

To draw a picture from scratch, if an AI is an intelligence, all that is required is 50-500K of code for brush movements and brush and palette generation code, and time. An online connection or offline collection of pictures is not required. So from this, we conclude that AI is not an intelligence, or at least cannot experience and gather impressions for itself.

To draw a picture from scratch, if a human is an intelligence, all that is required is motor skills for brush movement, brushes, colors, and time. We conclude that a human is an intelligence, because it can gather impressions for itself and express them in a painting, without an online connection, and without an offline collection of pictures.

The Demoscene can be the guardian of art, as it has been in competitions for decades. Copying, especially artwork that is not your own, is looked down upon, and if it's converted, the fact is quickly hidden by manual touch up (even when perhaps the straight conversion would preserve some of the manner of the originally painted picture!)

I think at any point in time, you will always have to choose between stuff and art.

We are already experiencing spam on the image front in various "web media", and also in demos. Streaming services are now dealing with "musicians" ready to be the kings and queens of the world! Of spamming.

For coding we are moderately safe so far, with a <1% success rate in producing working code. Bots only go by what's popular, i.e. top ten in web searches. Who will be the first to pay to make wrong code more popular than various coding sites? It shouldn't cost too much money, and could make AI steal incorrect "stuff". ;)

Alternatively, coding resources could go behind an account wall, go dark, or simply taken offline to not feed the bots (after changing the coding information to the wrong advice, of course!)

The same could be done for images, of course. Pay to promote images with humans having 3 legs and 1 arm to popularity, and AI images will reflect it.

As for the practical, low efforts will always smell like spam. Orgas can preselect and question authors of too good to be true things, and viewers can say what they like and don't like. Group orgas can question too-quickly-produced resources, and friends can say to friends, "Wow! that looks like photo, how did you draw that?"

It's a question of valor.
added on the 2024-04-11 21:24:29 by Photon Photon

login

Go to top