#because we don't get free choice of media. nobody will fund a cel animation film now.

[canmom]

How aren't mediums fungible? Any art history class would teach you they very much are.

what, has 'the medium is the message' gone out of fashion now or something?

but to explain what I'm trying to get at, since there's a good chance I misused the word >< - each medium brings its own set of affordances and emphases. if I see a CG animation I pay attention to different things than if I see traditional animation or stop motion or what have you.

for example, we could have a look at the animation of Hiroyuki Okiura - say, the introduction to the Cowboy Bebop movie, or his work in Magnetic Rose. Okiura is one of the most renowned realist animators, someone whose drawing style, camerawork etc. hews very close to live action film. his exceptional sense of perspective and space is remarkable in traditional animation. by contrast, you 'get it for free' in CG and stop motion - you will always have perfect linear perspective unless you go out of your way to break it. however, CG rarely captures the exact qualities of Okiura's animation, which come from the sense of drawing principles - how to simplify shapes, 2D spacing etc. and by making it something constructed, the way characters move through space, the way a drawing can suddenly feel 3D, becomes foregrounded - it's no longer incidental but now a core part of what Okiura's animation is expressing.

so, 'live action into anime' is kinda what the AI style transfer tools are going for. in the technique from the recent paper, you start with a static drawing of a character and some animation data (likely mocap), and the program will generate an animation. that's similar what Corridor Digital attempted a few months ago, using a neural network finetuned on Vampire Hunter D: Bloodlust, and applying 'style transfer' to live action footage they shot. the results were, viewed as rotoscoping, kind of hideous, with shapes constantly flickering and turning into mush. the new paper I linked offers some techniques to improve the temporal consistency of this type of AI rotoscoping which should make it look a lot less bad, though it remains to be seen whether it works in situations other than 'well-lit fullbody shot'.

still, even if Corridor's video was a lot more technically solid (and give AI development a few years to iron out the kinks, I'm sure it will look downright quaint), it doesn't provoke the same response in me as Okiura's animation. the process of drawing something involves a lot of artistic decisions about what to capture, simplify, emphasise; for all that it is 'realist', Okiura's animation likewise has a particular feeling to the way characters move, the way they interact with light, the use of line, etc. which in some large part arises from how it is produced. so much of that is all but impossible to capture in words.

but also - knowing a bit about how it's made, and having my own experiences of animation, gives me an angle to appreciate what Okiura is doing. a drawing of something is a way of drawing attention to the specific details of the subject. two people drawing the same subject will never draw it the exact same way. one of the joys of going to life drawing is seeing how many different ways people can approach the same subject in the same ten minutes - inflected by different media like charcoal or watercolour pencils. one of the great things about anime is the space it gives key animators to bring their own sensibility to a particular shot.

I certainly accept that is inevitable that mediums will evolve with time. anime looks very different today than it did 30 years ago. part of of that is evolving sensibilities, partly the slow-motion collapse of an overstrained industry, but also a lot has do with the fact that every studio has switched to digital compositing and digital background painting. it's possible through painstaking effort to fairly closely imitate the look of cel animation on a computer, but you really have to go out of your way, and it's rare to do that.

and I do feel like something has been lost with the death of cels - qualities of line and colour, the difference between digital bloom and backlight animation. but something has been gained at the same time: maybe we've gradually lost the traditional skills for drawing layouts because the conditions of production made it so that skills weren't passed on to the current generation of animators, which sucks, but we have simultaneously gained the ability to merge 2D and 3D animation with tools like Grease Pencil, to use the camera-like digital compositing effects of directors like Naoko Yamada and Makoto Shinkai. it's not better, just different.

this isn't to make the boring argument that AI art is soulless, or lacks the magic human touch, or whathaveyou. it's just a different medium. nor would it be right to say that there are no connections between media - literally right now I'm modelling an arm, and my experience of drawing arms is directly influencing how I break down the forms and all of that. AI generated images derive in obvious ways from traditional animation and CG and photography and all that, AI engineers study these media in great detail as they develop their programs; our knowledge of those media can inform how we respond to AI.

honestly, CG that aims to replicate the look of traditional animation, such as in the games of ArcSystem Works, or the works of Orange like their Houseki no Kuni, is something I actually find very interesting. not because I think it could or should replace traditional animation; it just reveals fascinating things about both media. the same can be true of AI, I think. like what do you learn from what a neural network is able to capture, and what it isn't? and what does studying neural networks tell us about human brains?

if the development of AI and the accessibility of new tools leads to a flourishing of interesting new animation, I'll be happy. I just don't see it as a replacement for traditional animation and 3DCG. if anything the future of animation will probably look like a hybrid process taking advantage of the best features of all the different media we've invented - insert the usual spiel about Arcane and Spiderverse here. AI is currently very immature, we're still figuring out what it's good for and the hype drowns out everything, but I'm sure it will find a comfortable place, and I'll be interested to see how it all shakes out.

but what I meant with 'not fungible' is that, if you try to replace one medium with another, you will inevitably change the qualities of what you make. nowt wrong with that. like, just because you can adapt books into films (and vice versa) doesn't mean books are obsolete. some things are easier to express in prose, others in film. you can have prose that's informed by film, and film that's informed by prose. everything's talking to everything else, it's great! but the tools you choose are meaningful, and interesting. not just an irrelevant detail to be swapped out when "superior" technology comes along.