And we are done gentlemen :-)... New text-to-3D AI...
polycounter lvl 9
...our 3d careers i mean :-):
https://www.youtube.com/watch?v=xrhubB4GQU4I thought we still had few years... but it appears...no :-)
(did you seem any better 3D generative AI (?)) ( me - no))
Gonna get that resume for Macdonalds application ready :-)!
EDIT:
Replies
If you’re building little kick-nacks to go in snow globes, then fear for your job! If not, chill out
It’s early days, so things are likely to keep improving. But as it stands this is still very much a tech demo, with no clear impact on 3d creatives.
I LOVE doing those :-))!
Also this : https://restofworld.org/2023/ai-image-china-video-game-layoffs/
Western studios are probably wiser and not going all-in because they know that a lot is at stake legally. This is actually quite fascinating to follow :
Utilising OS hasn't reduced the number of staff required at any of the studios I've been at - if anything it's created jobs in terms of managing the increased volume of content.
The real victims in the long term are going to be content farms (OS providers, quixel) and photographers/illustrators who work on a commission basis at the low-mid end
https://www.maketafi.com/
this effectively allows you to describe a preset you want to start from - I wouldn't turn that option down, particularly if I had a lot of Daz characters to make
I guess I need to specialize in something AI proof.
do people really consume daz3d porn? bonkers!
and yes - it's fundamentally just procedural generation based on statistical analysis. there's no intelligence, you get the illusion of intelligence from the scale it operates at.
to put it in that context.
perlin noise is a prediction of a value at a point in space.
it looks boring because its simple
it's fast because it's simple
ML system predicts a value at a point in space too
ML systems are complicated
ML systems are fast because they're pre-trained
Now imagine a world where your procedural generation didn't all sit on perlin noise and instead sat on something trained on real world structures that could also adapt itself to inputs .
before you get too excited, remember you live in a world where there's no money to be made off training that ML system
This is the largest 3D dataset
Its very shadily scraped from sketchfab and anything free they could find, so very sketchy
I saw the job posting, took a look out of curiosity but it obviously was a hard pass because I personally disagree with training AI and the potential for screwing over 3D artists and myself in the long run for a quick buck. It also looks like they are backed by investors and claim they have 3 successful companies under their belt on their site (but they don't say which). Reading into them further it just looked like they are trying to get investor interest by faking good 3D models from the Ai model so they can pawn it off basically to some other investors like most of these shills do.
Sadly, I know someone who took up this job offer to help with cleaning up models. They had an unpaid art test for the position and though they didn't try it because it made no mention of character art being needed, it shows the position is shady and doesn't really communicate what they actually want. Same with the salary range, but I imagine that is garbage as well because these people I don't value art, or the creative process to produce 2D/3D.
Here is the unpaid art test:
Not only is this test laughable but it shows they are trying to mimic clean topo, UV's and an overall well put together model that is usable in game. Pretty sad, and it depresses me that there are people who will take up these offers and help these things along. I get being down bad on money but come on, this stuff helps no one.
Now of course retopo is tedious. But at least the resulting asset can be handled/painted in real-time.
Overall, improvement in tech and processing power has *never* resulted in faster overall production times. And while some studios will keep chasing higher and higher fidelity, Nintendo will still be laughing their way to the bank thanks to successful games produced very efficiently, under strict hardware constraints ...
I think this would change the workflow to something like we see in Zbrush. Texturing for example would involve a lot more vertex painting because it takes advantage of the high vertex count and for areas where vertex painting just can't do the trick, texture painting could still be used. You could handle UV's similarly to Zbrush with a polygroup-like feature where you draw the groups onto the model, then unwrap it. Baking I'd imagine would largely be irrelevant since most people only use it to get detail from the sculpt to the low poly. I'm not a rigging guy and don't quite understand weight painting that much, but I would imagine this would benefit too if you could use vertexes for weight painting.
IMO character artists will benefit from this or maybe that's just me looking at this through optimistic lens unlike a lot of other people who get doom n gloom syndrome over every subject surrounding AI.
https://www.washingtonpost.com/technology/2023/07/16/ai-programs-training-lawsuits-fair-use/
"I think this would change the workflow to something like we see in Zbrush. Texturing for example would involve a lot more vertex painting because it takes advantage of the high vertex count and for areas where vertex painting just can't do the trick, texture painting could still be used."
By all means, try it - and you'll see.
It's really not as straightforward as you might think it is.
Vertex painting/polypainting on sculpts has been available for about 15 years now on the asset creation side of things - yet it is barely ever used for game asset creation (even if we take retopo out of the equation). I am not going to go into the reasons why that's the case (as this goes far beyond the scope of this thread), but as said by all means give it a try. For instance you could tackle a full character that way (sculpt + hardsurface modeling + only vertex color painting), then get it converted to a proper game asset (using automatic retopo for instance, to eliminate that step for the sake of the argument).
This has been attempted before of course, and for prototyping it absolutely works ; but when attempting to ramp it up to real assets that need to hold up in front of the camera, it just breaks in all kinds of ways.
Now whether of not generative 3D AI could bridge that gap, who knows. But if it does, and if developed in similarly unethical ways to that of the current generative 2D AI ... the end result would not be an improved workflow, but rather a straight up erasure of 3D modeler as a skill
Uhh I think you missed my point entirely. Of course such a method can't work today because computer processing power is simply too weak. My point is that when processing power gets improved enough (i.e. you can run a sculpt with a few million polys at 60 FPS), retopology will largely be gone. Our current workflow works the way it does because that's what our technology is limited to. With improved processing power, that will push the limits and affect everything, not just modeling, which will lead to new workflows.
Now I'm not a lawyer, but from my understanding the source of ethics here currently does not matter. The copyright office allows people to legally claim AI art as theirs if they modify it enough from whatever the prompt generator spits out. That would mean the copyright office is only taking into account ethics for the end result of that work, not the source from how it was generated. Essentially I think the copyright office is saying there is nothing that they can do disqualify something based on how it was generated and I'd wager its largely because its hard to prove this kind of thing unless said work is very easily proven to be ripped off of someone elses stuff.
Even with a legal ruling in place that declares AI generated works as officially being unethical and illegal, whats going to stop people from continuing to do it anyways? More so on that front, how you would prove something was AI generated? I think the cold truth is we're going to have to learn to live with generative AI being used unethically regardless of what legal ruling gets made in our favor.
But, sources for game assets are generally not created that way, even though they *could be*, today. And the reason why the creation of sources assets isn't done that way isn't because game engines are not powerful enough to use such assets directly. The reason is that such polygon soup assets simply don't hold up up close (+ a few other reasons). That's why I am suggesting that you try and do it yourself, on a full, animated character. You'll soon realize that even if nanite or similar tech automagically allowed such polygon-soup models to be used in games, they would likely still not be created that way ... because it simply doesn't look good beyond renders with very controlled lighting. And furthermore this throws a wrench in the animation pipeline which *does* require cleanly built topology. That's not going to change anytime soon.
I understand that this isn't intuitive or straightforward to understand. I've actually met very clever programmers 10 years ago who where 100% certain that "normalmaps will soon disappear" because "everything will just use displacement soon". The assumption you are making is somewhat similar, and comes from a misunderstanding on how and why game assets are built the way they are, for both visual and technical reasons.
-----
Not going over the rest (legal/ethics of regurgitative AI generation, and whether or not companies working on generative AI tools will still be able to obfuscate their source datasets in the future), as this has been well covered already in countless places. I am simply saying that if AI-generated 3D content becomes a thing in the near future (even if ethically sourced), modelers will not have the luxury to enjoy it as a "tool to improve their workflow" for very long, as it would directly lead to them being not needed anymore.
In that sense I tend to agree with the main topic of the thread ("and we are done gentlemen"). If anything, if generative 3D AI ends up growing similarly to 2D, the only jobs left for modelers would be ... clean up work/retopo, ironically enough.
It's mindblowing really, as the alternative is right there in plain sight : Minecraft, Roblox and now Battlebit are insanely successful. IMHO It would be very logical for more studios to fully embrace proactively limited art styles and lean content creation. As a matter of fact it seems like it is already starting to happen outside of the indie world (BoltGun comes to mind) and that really is a breath of fresh air in these times of AI-generated vomit.
[edit] : I just found out that a sequel to Iron Fury (the Duke-like retro FPS) is coming, and it is using a very rarely used artstyle nowadays : 2000/2005 gritty handpainted textures, and it looks incredibly authentic. I often wonder what games would look like today if high end baking never happened, and this is a really good example of that.
I find that this is quite relevant to the topic of this thread, as this to me seems like a very healthy alternative to the "let's do AI !" and "let's Nanite everything !" trends.
This is where a lot my anxieties (concerning ai+art) are. While we're not there yet, it also isn't some far-fetched fantasy to imagine that that could be a possibility.
And to a similar idea regarding topology, UVs and workflow:
These systems were all developed primarily from the standpoint of human authorship, right?
How wild is it to think that some time in our dystopian future, today's AAA hyperrealism will be a "low-res" aesthetic style
I mean, how fun must it have been to work on BoltGun or that Iron Fury sequel ! And I can't help but imagine the dreadful atmosphere in teams where gEnErAtIvE aRtIfiCiAl iNtElLiGeNcE has been embraced, by contrast. No possible art knowledge sharing between team members, no feeling of accomplishment ... just nothing.
I just started a new playthrough of MGS Peace Walker recently and it really hit me - despite its lowtech simplicity, this game is so incredibly pleasurable to play and so beautifully crisp and readable. Even the raw and simple pupeteered cutscenes contribute to the suspension of disbelief.
I really wish for more visually similar games (barebones simple tech, but crisp visuals) to be made today. If anything it only benefits budget but also gameplay, since less feature creep gets in the way.
- - - - -
15+ years ago or so, some Polycount members were having fun doing HQ 3D remakes of Joust character sprites by the standard of that time (2000-2005 ish game art) with the hope of developing a sequel. Improvement in tech flew by, but not in a weird way I find that this careful crafting of tight, low-tech game art could be more relevant than ever. So perhaps not all hope is lost after all
You seem to be approaching this from the standpoint of 'but that ain't how it works today so your wrong', which is faulty nor is it respond to the point I was making. It doesn't really matter how games are made today. I'm talking new technology coming out tomorrow and beyond that will change how games are developed, possibly in a major way. What is so hard for you to accept or comprehend about that?
I understand that this isn't intuitive or straightforward to understand. I've actually met very clever programmers 10 years ago who where 100% certain that "normalmaps will soon disappear" because "everything will just use displacement soon". The assumption you are making is somewhat similar, and comes from a misunderstanding on how and why game assets are built the way they are, for both visual and technical reasons.
Both paragraphs are wrong from top to bottom. Your whole response is built on the false assumption that you think I don't know what character modeling is or how games work.
Many 2D artists also had a similar mindset about technology in the form of AI not being able to replace what they do, until they got proven wrong when the AI proved it can do art to their level. NVIDIA for example is showcasing groundbreaking changes in game development as it relates to 3D/graphics technology, perhaps not on the modeling front yet as far as I can see but it will get there.
All I'm saying is, for all we know in a few years we COULD have new groundbreaking technology that completely changes how games are developed. Dunno if that meets your approval of "soon", but don't be surprised if you get proven wrong.
Imagine dedicating your entire existence to being an art professional, just to realize that your higher ups don't GAF and you get a nice fat juicy 2.9 (and still dropping, as of now) on Metacritic. Or imagine making assets for Overwatch 2, only to have it be canceled but since your art is still under NDA you have nothing to show for the years of hard work. This industry at the higher end is a clown act and a half due to poor decision making.
While we're here busting our balls and nitpicking every facet of good topology flow, proportions, animation, materials or lighting, the executive class is lazy and incompetent, and we should be replacing them instead of them, us.
It's getting more obvious that authorship and uniqueness are the only things that are important and get results. Producing a bunch of generic current day 3d models that fit particular technical specifications is getting anyone nowhere. It's not even working for big studious now.