Forum adverts like this one are shown to any user who is not logged in. Join us by filling out a tiny 3 field form and you will get your own, free, dakka user account which gives a good range of benefits to you: No adverts like this in the forums anymore. Times and dates in your local timezone. Full tracking of what you have read so you can skip to your first unread post, easily see what has changed since you last logged in, and easily see what is new at a glance. Email notifications for threads you want to watch closely. Being a part of the oldest wargaming community on the net. If you are already a member then feel free to login now.

Hello all,

I was ruminating on the future trajectory of A.I. art recently, and imagined the following scenario:

What if you could tell your subscription A.I. program something like, "I want a selection of 100 Eldar Striking Scorpions, sculpted in the style of Jes Goodwin circa 1988, with a touch of John Blanche-style ornamentation, in samurai poses from Akira Kurosawa's films." After the program instantly generates 100 examples, you'd then pick the best examples out of the 100, or maybe ask for some specific revisions. Then you'd print them.

Assuming the A.I. program was capable enough, I think that would ruin the job market for human sculptors, whether they sculpt by hand or digitally. I'm assuming that for most human artists in this business, their margins aren't fantastic to begin with, so even if A.I. only captures a portion of their business that could be enough to make their work untenable. A few short-sighted companies might celebrate at first, thinking "Great! We don't have to pay artists anymore!" But with A.I. design and 3-D printing accessible to the average middle-class household, there wouldn't be much need for miniature companies, either, whether they're selling physical products or STL files.

If this actually happens, some people might argue that the change is "liberating" and "makes us all artists." For me it would suck. I value art and design made by humans, but I tend towards pessimism, so I tend to think that the vast majority of miniature gaming customers would be fine with their A.I.-designed 3-D prints.

Hopefully I'm completely wrong about all of this. If I'm right, then we really are living in the "golden age of miniature gaming" right now, and that golden age isn't going to last long.

AI can only aggregate based on existing material. Original material will be in the hands of the creatives for a very long time. If AI > 3D printing is an issue, then the solution is strict enforcement of copyright.

Generative AI works by illegally scraping content from the internet. Given that 3D models are basically nonexistent as something that can be scraped en masse for free (and the quality of those that can be is abysmally low), we don't really have to worry about generative AI doing much to the sculpting field.

I've heard that in other areas of AI it's reaching a recursion point where AI is scraping so much AI generated content the results are degrading. A copy of a copy of a copy if you will.

It's definitely not working out to be the miracle/horror it was initially proclaimed to be.

 ScarletRose wrote:
I've heard that in other areas of AI it's reaching a recursion point where AI is scraping so much AI generated content the results are degrading. A copy of a copy of a copy if you will.

It's definitely not working out to be the miracle/horror it was initially proclaimed to be.

The AI models that stick around long term will be those trained on specific datasets, like Adobe's which is trained on open source images and those they specifically own the rights to, rather than those that scrape the internet.

Miniature sculpting though feels like far too much of a niche to warrant AI development... and given that many of them are still struggling with generating the right number of fingers or teeth on 2D images, I wouldn't expect AI to be capable of producing functional 3D models any time soon.

Not a direct answer, but...

Specifically for GW, we already have almost-legal stl's to print GW miniatures, so the "not paying GW to play GW" is already a thing. I mean, there's even a Reddit subforum for printed WH stuff. Even without a 3D printer, you can purchase "alternative miniatures". : https://spikeybits.com/2022/10/50-alternative-companies-to-games-workshop-to-try-in-2021.html

As for AI creating new miniatures, HeroForge has already made free software that lets *you* make your own miniature. You know, like how AI lets *you* make your own miniature, but without the hype. Obviously, this is for original work (for now), but it takes a sculptor out of the equation. So who needs AI when you can take away someone else's job all by yourself? And, yes, there's a reddit forum for HeroForge, where people are *giving* away their creations.

That said, when GW said enough was enough with Warhammer Fantasy, it replaced it with the more IP-able Age of Sigmar stuff. Then Mantic's Kings of War and other companies that said it was okay to use your own miniatures appeared. Interestingly, that didn't stop the market from not just buying official Kings of War miniatures, but Mantic now has a 3D print subscription. Good work, GW.

Really, the root problem is that all digital content (okay, not NFT's) can be copied, and we don't have a solution to prevent it. Yet somehow the market can support digital content, entire industries, such as television, movies, music, artwork, and so on. While a new technology can consume a fair share of a previous technology's market, that doesn't mean it will be an "end of it all" scenario -- sometimes -- and often a new technology will make things cheaper and increase the market.

Can't see it happening for some time.

I'm getting back into D&D after a long interlude and have been looking at miniatures, including 3D.

For me, far too many items on sale are illustrated as renders, which give me no clue what the final article will look like. Of those printed, still too many figures are blurry/smudgy, or have obvious print lines,. like a mini lego ziggurat. There IS quality stuff out there, but OP's scenario seems a long way off form this desk...

3d digital sculpting churning massive amounts of cheap virtual sculpts and turning traditional sculpting unwelcome, will face a taste of their own medicine, cheaper unregulated and more canibalistic methods. Most digital sculptors not even from a miniature background so they dont care much either way.

What does that mean for consumers... cheaper and even less quality control.
I will stick to who actually produces miniatures in the traditional sense of holding physical stock and selling.

From what I've seen of AI doing its own stuff from scratch its basically super unreliable. It can make something awesome, but its important to realise that all the AI is doing is taking key words and producing a collage of the results.

In theory you can break it just by having something abnormal keyworded with words that don't describe it correctly. Eg if enough people keyword a picture of a dog with cat then AI could pick up on that.


The other thing is that because it doesn't know what its doing, it will produce random/wrong results.

This means you might have to run it many times to get a good result; or have the key creative skills to go in and fix what the AI did to start with. Meanwhile the more complex the creation the more time you'll have to devote to fixing what the AI broke, with a high chance that very complex projects might end up with a lot of errors that you don't spot until later.

At which point its likely easier to pay someone to do the work from scratch than it is to use AI to get close and then finish up.



So far I see it being used more in situations where those using the AI don't have key skills (or can't afford to pay for them) and are thus happy with "gets the job done" level of production.


The other avenue of course is creatives using AI elements in support of their own work. So they might not use the AI to sculpt a dog, but they might use keywords for the AI to sculpt parts like ear, claw, paw and such to then use as a creative inspiration and such.










That said AI is scary, but its legal status and the legal status of what it makes is in question when it comes ot copyright. I also think there's a genuine need to socially discuss the element of what if it results in a reduction in the viability of creative work resulting in a reduction in jobs and loss of key creative skills.

Laughing Man wrote:
Generative AI works by illegally scraping content from the internet. Given that 3D models are basically nonexistent as something that can be scraped en masse for free (and the quality of those that can be is abysmally low), we don't really have to worry about generative AI doing much to the sculpting field.

There are an awful lot of 3D models out there in games that could easily be the basis of anything you want. There are also an awful lot of really good quality models on free sites like Thingiverse. If anything, the AI doesn’t need to know how to sculpt, but just reformat existing 3D models intended for other purposes into an easy-print file.


Automatically Appended Next Post:
I would also say that it potentially gives the power to individuals to create something on the scale of 40k in a fraction of the time. The issue then is more about fragmentation of the player base as everyone literally makes their own thing.

Video games models are A) stored in formats not easily usable or readable, and generally require specialized software to rip them to something that you can modify, B) are extremely low detail, instead faking details via textures and rendering tricks in the engine, C) are stored in a T-pose that will make your AI-generated army extremely interesting, and D) are completely unprintable due to a combination of holes in the mesh and storing every conceivable part for the model in the same model, choosing to make those parts invisible when rendered and not needed.

Thingiverse, meanwhile, has anti-DDOS systems in place that limit how quickly models can be downloaded, which is going to prevent an AI from trawling the site for models to download. Files tend to be absolutely massive compared to trawling for images and text as well, which increases the cost to train an AI like this dramatically. Thingiverse free models are also nearly all very low quality, and your AI is going to be trained on the flood of gak on the site.

Finally, training on game rips and Thingiverse downloads alike is incredibly illegal. Game models are obviously copyrighted by the company that produced them, and Thingiverse models all are published under a Creative Commons license that (by default) forbids commercial use or derivatives.

I suspect Imperial vehicles and then marines could be done by AI reasonably easily, firstly because the vehicles are boxy and use a lot of copy & paste parts and secondly some marines look hidious by design

T-shirt cannons anyone?

that and a fair bit of GW design while intricate seems a bit lazy with copy & paste as well

it will be a while yet but probably not that long before "I want a space marine in a kneeling pose with a bolter" is possible

Things that have sounded the death knell for the 'Golden Age of Miniature Wargaming' of the last 40 years.

Additive Manufacturing/3D Printing/Patreon.
Sugru.
Green Stuff Moulds.
The common availability of Green Stuff and other epoxy putties.
Disney.
Amazon.
Ebay.
Google.
The Interwebz.
Computer Gamez.
Satan.

to be fair on 3d printing, I bought a FDM one last year and a resin one a week ago, there is an upfront cost and a learning curve without question

however a Deimos Rhino seems to be about £3-£4 in resin, and can be printed pretty quickly.

this won't kill GW, but it will start to bleed them, they still apparently see their greatest competitor to be eBay, party resulting in updating old models (which I'd say is usually a win all round as they tend to be a lot better)

 ced1106 wrote:
As for AI creating new miniatures, HeroForge has already made free software that lets *you* make your own miniature. You know, like how AI lets *you* make your own miniature, but without the hype. Obviously, this is for original work (for now), but it takes a sculptor out of the equation. So who needs AI when you can take away someone else's job all by yourself? And, yes, there's a reddit forum for HeroForge, where people are *giving* away their creations.

HeroForge works by using pre-existing digital sculpts made by a human that you then put together how you choose using their application. It's not the same as AI art and doesn't remove a sculptor from the process.

@Laughingman - if a party is writing some kind of code to automate model creation, then your a to d points are irrelevant, as there can be software modules To deal with each of them.

I guess it comes down to the intention of the AI or automation widget. If it is to enable full armies of specific IP related models to be created, then you don’t need to train it on that much. One decent set of armour/uniform plus as many random widgets as you want, then what you are actually training the AI to do is generate cool looking poses using a specific set of base models.

If you want to be able to generate armies from scratch then it’s probably more about training on the body form that you want to model, plus a broader range of armour and weapon aesthetics.

You absolutely can't train an AI to build a model and then pose it. Doing so would require the raw blender files for the models' rigs, which virtually no vendor is going to give you, let along free models on Thingiverse. You also aren't going to be able to overcome game models being a low poly unprintable mess covered with extraneous details floating in midair and overlapping each other.

Ripping software also is typically specialized to a single game, often costs money, and requires you to have the game installed for it to even work, and again, ripping games doesn't yield a useable result without a massive amount of work by the ripper.

And again, all of this is extremely illegal. The moment someone even tries to train a model for this sort of work they're going to get sued into the ground, making Getty's lawsuit look like a friendly hug.

Even if AI could design a passable STL for printing, you'd still need to kitbash the model new hands and feet

Laughing Man wrote:
You absolutely can't train an AI to build a model and then pose it. Doing so would require the raw blender files for the models' rigs, which virtually no vendor is going to give you, let along free models on Thingiverse. You also aren't going to be able to overcome game models being a low poly unprintable mess covered with extraneous details floating in midair and overlapping each other.

Ripping software also is typically specialized to a single game, often costs money, and requires you to have the game installed for it to even work, and again, ripping games doesn't yield a useable result without a massive amount of work by the ripper.

And again, all of this is extremely illegal. The moment someone even tries to train a model for this sort of work they're going to get sued into the ground, making Getty's lawsuit look like a friendly hug.

It wouldn't require ripping anything. All that would be needed is an AI that can turn 2D images into a 3D sculpt. You enter your prompt, it creates a bunch of 'concept' images, you select the one you like and press 'Print' on your 3D printer.

That 'all' is still huge, however, and certainly well beyond the capabilites of the AI being played with in the wild at the moment.

Laughing Man wrote:
You absolutely can't train an AI to build a model and then pose it. Doing so would require the raw blender files for the models' rigs, which virtually no vendor is going to give you, let along free models on Thingiverse. You also aren't going to be able to overcome game models being a low poly unprintable mess covered with extraneous details floating in midair and overlapping each other.

Ripping software also is typically specialized to a single game, often costs money, and requires you to have the game installed for it to even work, and again, ripping games doesn't yield a useable result without a massive amount of work by the ripper.

And again, all of this is extremely illegal. The moment someone even tries to train a model for this sort of work they're going to get sued into the ground, making Getty's lawsuit look like a friendly hug.

Extremely illegal as in completely?
Or just if you do it to someone else's work without thier consent?

Training an AI to recognise the human form and create skeletal poses doesn’t seem illegal. Training an AI to recognise the basic forms from 3D models (I.e. basic limb and joint identification) doesnt seem illegal. The vendor can’t control what the end user then uses it clothe their new doll-maker software. If the user of the AI only uses it to create models for their own use it may even fall under permitted use cases for models on Thingiverse.

In order for there to be a miniature apocalypse, it would be necessary for the training data set for "AI design of 3D models" to be feasible to build.

It's not, because 3D modeling is divided among numerous subdomains based on application. ThingVerse STL files don't contain the same sorts of data that 3D game files do--each has different data that the other considers extraneous but is vital for the individual uses. The stl file you feed into your slicer to get a printed model doesn't have the data needed to change the model's pose, for example.

Compared to 3D, 2D graphics may as well be using a homogenous data standard. That homogeny is how you can have AI trained on large enough sets of images to produce useful results.