In my mind, too, though I’m biased for obvious reasons.
From a competitive standpoint in creative jams, AI has, to me, the (completely unscientific) feel of giving (if used in large amounts) the same sort of edge as steroids or PEDs. But I can see how people might think, “why shouldn’t we use the best tools out there?”
However it also has the feel of using autotune.
And I think autotune overall has been bad for the pop music industry. But the one bright side of autotune is that there are all sorts of channels and such on YouTube where people enjoy how good older artists’ voices are and they hear the originality and character etc. even if they can’t verbalize it and that means something to them.
The doomed Great Mouse Utopia project proved that UBI doesn’t work. It violates “A day without work is a day without eating” principle. I think you can see how “free” living works in CA and other places.
I was tired when I wrote my last comment, and don’t want to restart the contentious debate about AI images. I’m not tired now so I don’t have that excuse anymore, but I’m not sure I can help myself.
I am ambivalent about economics, technological progress, and the organization of society, which are all firmly outside my ability to influence or control, and I find it generally unhelpful to have too many opinions about them beyond the purely pragmatic.
However, I think there’s a reasonably good argument against the aesthetics of AI images. If you don’t have the skill or interest in creating images, you should do those things you do have the skill and interest for. Only through that will you be able to express your unique perspective and find connection with other people. Montaigne discusses the phenomena of pseudo intellectuals who claim to have big ideas but who lack the skill to describe them. AI is not like a new kind of paint brush - it is more like giving your big idea to a computer and telling it to express it, and you will get the computer’s (soulless) expression, not your own.
personally, im not against AI in games, but for the art, i think it’s going to be some time before AI can make suitable art for games.
In a game, you have various lead characters. you need these people drawn in different situations, doing different things, with different expressions on their faces.
Right now, i dont think AI has the context to say, “draw larry driving a car” now “draw larry climbing a ladder” now “draw larry chatting at the dinner table”.
People tell me, it will be able to do that soon. But personally, im skeptical. And even if it could do that, you’ll find that suddenly the cost of such AI will be high. RIght now these people are charging $30/month (or so). If it really worked, then it might be more like $99/month. And that’s already out of most indie budgets.
People have been claiming unrealistic stuff about AI for half a century. It’s always ‘just around the corner’, like practical fusion power.
I play around with AI once in a while and everytime I come away with the same impression: It’s a lot of smoke and mirrors. The language stuff is dangerously inaccurate in a ridiculous number of cases and anyone relying on it is a fool. The industry’s marketing has even given AI’s propensity to be flat out wrong a cute euphemism: hallucinating. In reality it is just more obfuscation to promote an impression of intelligence when the reality is the ‘AI’ has no idea what the words it is spewing even mean.
They were banging on about AI taking everyone’s jobs back in the '80s. It was going to replace factory automation etc. it’s basically the same thing today, but on faster computers.
There’s a video on YT where Geoffrey Hinton admits AI is 17000 time faster than the 80’s, then he explains computers are the 1000 times and the rest is small improvements.
Well, following my argument above, the more capable the AI is the less I’m interested in it. I am not interested in art from the perspective of a computer. Maybe in the future that will change, but right now I’m happy being prejudiced against them.
You can absolutely do this using existing AI tools, it just requires a fair amount of technical expertise (ControlNet, Photoshop, digital painting skills, bespoke model training, etc.). Plus, you know, artistic vision.
I don’t think the objection is primarily to AI art in general (except for those who have deeper philosophical problems with AI), but rather to lazy AI art that doesn’t amount to much more than illustrating your game with lightly customized stock photos.
I’m not sure I’ve thought about AI deep enough to have deep philosophical problems with it. My problem is more that when I look at a picture and try to think about what it means and how it makes me feel, I don’t want to waste my time and energy on it only go discovery that there is no meaning, only a void. There is too much actual good art in the world to be distracted by computer generated art.
Can you show me two images of the same character in different poses. Let say, standing (which should be easy) and, say, sitting (which should also be easy).
For those interested. This is the indomitable Zulin from my forthcoming game “The Italian Maxman and the Alien Invasion”. A sci-fi comedy!
The game also features Brian Rushton as a disembodied brain… And Chapter 5 is written. Brian doesn’t get unplugged after all. Zulin saves him… Y’all gonna have to wait to find out what happens…
You can’t do this with prompts alone; there’s a whole toolchain of different techniques, particularly ControlNet (long tutorial with many demos).
For some reason I can’t quickly find an example that includes sitting specifically, but this is the sort of thing you can do (credit):
Getting really good character consistency is still nontrivial, though. Expect to spend a lot of time tinkering your workflow, generating variants, inpainting or compositing together the best bits from multiple runs, and using supplementary techniques like face-swapping AI or training LoRAs on reference images of the character you want. Although you can get a long way just by blending together ideas the model already has a good idea of: for example, the experimental visual novel Zarya of the Dawn (subject of a recent high-profile copyright dispute) achieves consistency mostly by making the main character look uncannily like Zendaya.
I often point to Stelfie the Time Traveler as an early example of AI art with a consistent character and real artistic vision and merit, as well as the depths of effort you can put into AI art beyond just promptwriting. (*note that tools have gotten a lot better at photorealism and body proportions since the linked video was released.)
I had a lot of problems with AI getting the whole body-shot. I tried specifying the boots in the prompt, but even then it would sometimes not draw them. You’d just see the tops of the boots and not the toes.
