AI Art was all the buzz a couple of weeks ago. That chatter has died off somewhat, perhaps as people got tired of the shiny new toys like Midjourney and Dall-E 2, but it’s a Thing that will only get more technically impressive and practically useful as time goes on. The pros and cons of that can certainly be debated, but I don’t think that we’re going to see that genie go back into the bottle. Like “Machine Learning”, which improves things that the Money Men care about in production, like schedule and headcount, using AI in art is a tool and a toy that is too useful to go away. At least, until the inevitable meltdown of society and technology, and we’re back to drawing on stone cave walls with charcoal-tipped sticks, but that’s tangential.
“steampunk floating island apocalypse” via NightCafe
This particular bit of buzz is of interest to me both in the abstract and professionally. I worked in video game development for a decade, and I’m working in film at the moment. I haven’t had occasion to use these particular tools for anything more intense than helping my kids with homework, but I do use Houdini, which is built on “proceduralism“, which is more or less the engine that drives AI art.
I’m already using a tool that takes inputs, runs simulations and variations, then spits out something that I can sort-of art direct. The computer does the heavy lifting of calculating all the bits and bobs bouncing about, and if I’ve set up the parameters for the procedure correctly, that calculation comes up with something usable. My job is then mostly about setting up the system for success, and inevitably wrangling things when the computer mangles them somehow. I’m not drawing and painting frames, like I grew up wanting to do, watching the Nine Old Men work their magic. No, I’m a desk-jockey cowboy-mage, desperately trying to harness eldritch powers in a digital wilderness, hoping to produce something that the art director will be happy with.
I’m using a tool to produce effects. It’s not the same as using a ballpoint pen on paper, which I can do, as seen here with my Dwarven Tinkerer, but it’s still a tool. It’s a tool with a bit of a mind of its own, and a black box heart that I hope I can channel to great effect. Sometimes it does as predicted, but sometimes it gets a bit flipped somewhere, or an assumption inverted, and things go awry. This, to me, is the most irksome part of using such tools from a production standpoint. Yes, the simulations get faster and faster every year, the results cleaner and more useful… but sometimes I just don’t have the control that I have with much less ambitious (and much more time consuming) tools.
Maybe I can have the spiffy AI system generate 200 different trees, all variations on a theme based on growth rules and parameters, but none of them are what I actually want to use for a “hero” tree. They can be good for fillers to back up the Potemkin Villages that games and films build as part of their magical facades, but for things that get the spotlight, that Uncanny Valley effect where computers still don’t quite get reality is still a hurdle.
We’ve known this for a long time in film; that’s part of why filmmakers can get away with matte painted backgrounds and greenscreen tricks, even as they spend an inordinate amount of time on actors and their makeup and lighting. Backgrounds can be simpler, counting on viewer assumptions and interpolations to gloss over imperfections. We also see a similar “audience interpretation” filling in the gaps when we look at concept art. Even masters like Daniel Dociu, for all their incredible skill and intricate detailing, still don’t work out and carefully render every little detail when they produce concept art. Zoom in on something like his “Tectonic Dystopia” piece…
…and note that even as he bombards the viewer with detail, it doesn’t always bear heavy scrutiny. He’s put in a lot of work, but a detail like a single road is largely a suggestion, a brushstroke or two, maybe a few blobs or smudges, and the viewer’s assumptions of what a city looks like at scale fills in the mental gaps. It’s a fine dance between just enough detail to be plausible without having so much detail that it triggers our sense of wrongness if something’s not perfect.
Leveraging the viewer’s imagination and interpretation is indeed part of Dociu’s mastery of his craft, and while I may sound disparaging, I recognize and am impressed by his genuine skill in performing such feats. Sometimes, we want to be fooled. Art movements and forms of entertainment have been built on this sort of shenanigan, tricking the viewer’s eye, like pointillism, impressionism, or the mental assault of cubism and lesser imitators in more modern art, bluffing with balderdash to give the impression of depth.
The principles at play, then, those of dazzling with detail, or overloading with obfuscation, well, those are age-old fine art traditions. When it comes to AI, though, it’s still learning. It’s only as good as the material it’s trained with, and the assumptions built into the generation systems. Those assumptions aren’t always built with fine art principles in mind, or are built to function first, rather than consider fripperies like composition, emotional appeal or verisimilitude, much less photoreality. Perhaps such considerations will continue to be folded into the frameworks of these tools, but for now, there is a lot of room to grow.
Deep Fake videos are one branch of the technology that is getting particularly interesting and potentially troublesome. Sure, being able to fake Tom Cruise or Harrison Ford is a humoresque parlor trick, but more nefarious uses abound in an era of political disarray and general lack of fidelity to truth. There’s a moral dimension to art, and there always has been, so it’s wise to be aware of how technology can engender trust when it is not warranted. Again, sometimes people want to be fooled, though, for better and worse.
Similarly, there are revolutions in animation brewing. Motion is especially tricky, and much more likely to faceplant into the Uncanny Valley. The technology keeps improving, however, as noted over here, and here. This will definitely make some production faster, especially for midground and background crowds and such. It will be interesting to see how well it fares in the foreground. I’m not convinced yet that it will work as well as some would like, but there are already real consequences for production pipelines.
In the meantime, however, I’ve found that I increasingly value authenticity. From OK GO‘s oddball music videos that bank on their intense efforts in production to Wintergatan’s fascinating machine, from anachronistically authentic YouTube gamers (the older gentleman known as TinFoilChef just played Minecraft the way he wanted to and built an audience that loved his affable curmudgeonly ways) to hand-carved woodworking, I find value in things that appear to me to be genuine and honest. I still carry pens and a sketchbook most places I go, after all, and I’m almost always drawing something, even if it’s just odd designs to keep myself focused. There is value in things that have been made by hand, though whether that value can translate into a career is certainly always a question.
The sky isn’t falling. New tools mean more ways to fool people, with all its attendant implications for an increasingly dysfunctional humanity. “All is vanity“, though, and we must always consider truth and our own decisions. It was ever thus. My profession is definitely impacted, and my personal interests in creative endeavors will be perturbed somewhat, so I’m not neutral on this. I simply see yet another set of shenanigans. Artists have always borne responsibility to be uplifting and useful, since their tools are inherently not honest, as mere representations of reality. Far too many fail miserably at this, and new tools will not compensate for moral failures. Those of us in the audience will always have to be wary, or at least, we’ll have to choose which artifice we want to accept as authentic.
Tish Tosh Tesh 