Ubisoft has basically been doing that in their tour modes for Assassin’s Creed games lately, and Watch Dogs 3.
I think the people most involved in arguing against the use of AI tools to produce things like this would probably never be hired by these companies to do that work anyway, only because these companies seem to hire people mostly proficient in getting hired and practicing a styleless craft instead of actual practicing artists. I’m also really sympathetic with the surrealists and find nothing about AI art production strange or scary in any actual way. If anything, when artists online complain about AI works being intrinsically valueless compared to work produced by a human I kind of think they’re drawing the wrong conclusion about this.
It’s coming, we were investigating the voice synthesis tools years ago and were under the impression EA and Ubisoft were much further ahead. The image tools are kind of shocking but they’ve been extremely useful in both of my projects this past year, in teams with full art staffs. In production, they’re enormously useful for concepting (“give me another 20 riffs on this idea”) and prototyping, expanding what each artist can do.
On the other side, we’re looking at production use for some of this and thinking about what it could enable: new types of procedural generation, a step magnitude in asset count, an entirely different approach to asset conservation. It’s not clear to us yet how valuable it can feel to the player when they have more-customized, more-unique illustrations of procedural outcomes that they were already invested in and which we’d normally represent with reused, generic text and images. I’m not satisfied with reading AI text on the internet but I’m coming in with very low expectations; on the other hand, I’m very investing in what gets churned out as I’m looking for specific images with specific ideas. I’m curious how it will feel to a player working with characters they care about.
I think that’s what we’re likely to see in final game assets several years down the line. I expect to art teams use it to fill out world assets by working on the style and building out hero objects, training a tool on their style, then using it to fill out variations of tables, vases, and small material culture.
i don’t think this stuff is surrealism, i think it’s the opposite of surrealism, it’s training a machine on max ernst collages to eventually spit out the schmaltzy woodcut sources of une semaine de bonte - the kind of professional art which ai is pretty interchangeable with is already aimed at turning people into machines of ossified technique and best practice, but there’s always at least the possibility that shit might be pulled off course by the gravity well of the artist’s own desires or inattention or the refraction of their thoughts through some material, whereas the generated posters in bachelor’s post are so dull because any deviation from commercial template just manifests as a kind of evenly distributed stochastic fuzz. i think this stuff will be videogame worthy the moment it’s able to come up with posters as funny and strange as the one in Mappy Kids and not before.
In certain ways, it’s like a search engine, and the types of things that people are spreading are double-concentrated: searching inside the latent space of popular image styles. The most popular popularizers are the least tasteful.
Or, the internet at large is using it to create internet-style culture mashups, of which the internet already had an infinite supply.
It’s not a limitation of the tools. An interesting thing about the tools though is that it lets you ‘feel’ the density of information on the internet by the quality of results. Denser nodes return higher-fidelity results, and certain surprising things aren’t well-understood. For example, it’s hard to get people eating food, because so much visual culture is pictures of food alone. (it’s also the case that the doubling of fidelity every 6 months might squash this artifact out)
AI Art I think in general does look kind of distinctively lame, its neat as a process but every end result is transparently deficient & thats clear from the fact that they couldn’t even sneak it into the pastiche rick and morty 80s game without people noticing. This seems like the biggest issue to me, maybe I’m being pessimistic here, but I think overall AI generation creates things that are all stylistically similar in a way that is clearly identifiable (especially once you’ve seen a few), and almost entirely reliant on novelty & convenience (check out the prompt!).
I think it’s pretty easy to tell when you’ve stumbled into an SEO website and the more AI processes we see snuck into products the more I’m sure the same will be true for whatever function they end up serving in a workflow. Personally, I do find like ChatGPT blabber kind of charming because it comes off to me as political and avoidant. I don’t think it’s impossible to do something interesting with these tools taking in mind their obvious deficiencies.
But all this automation clearly has its main function in offering a justification for cost-cutting at points where a real human being would need to be hired. Even if it’s only used at conceptual stages, it’s obvious that whatever qualitative difference you do/don’t perceive, a seemingly cheaper & easier path is gonna appeal to those footing the bill (and collecting the checks). Outside of the larger point that less demand for artists means less people pursuing art, or at least, less who aren’t already financially secure, etc… It’s hard for me to imagine how a reliance on these tools doesn’t lead to a sense of homogenization even if it’s hidden away. Everyone’s hiring the exact same forger to assist innovation…doesn’t seem ideal.
E: I see most of what I had to say was said better but I typed all this out so I will stand by it
I don’t think it’s surrealism, but I do think a lot of people that feel threatened by this stuff because it supposedly devalues their own work would also have the same resentment about surrealist methods of production. A lot of people don’t understand how mundane AI art production really is. It does nothing to threaten or devalue art made by humans any more than what corporate logic and professionalization already have accomplished.
we’re already well past the point where there are successful artists who weren’t financially secure beforehand. Arts and literature in the 21st century is composed of exactly the same class of the idle rich as was the case in the late 19th century. Conditions are certainly worse now than they were a century ago (given that average illustrator pay, adjusted for inflation, is something like 8% of what it was in the 1920s)
We’re really caroming to the point where information truly is valueless because it’s in infinite supply, and without an economic structure that can plausibly place a value on that work.
it’s not just that it’s pictures of food alone, but that we have tagged pictures of food by itself meticulously on sites like instagram. The LAION database is pure garbage-in as a method of database building. The misogyny inherent to things like Lensa turning any picture of an asian woman into a naked anime girl comes directly from the LAION database having scraped all those meticulously tagged anime porn image boards.
This is also why so many artstation artists are represented, not because the people making the database were planning on stealing art, but because the database is a sort of massive shortcut that obviates the need for them to make an ethically sourced and hand trained semantic model of text-image pairings.
That’s what I mean by feeling the patterns of data – it’s a unique experience to get a sense of the shape of something as massive as the internet in a non-numerical fashion.
I don’t think of the database as a massive shortcut as the biggest advances have come from each step-change in data size. So far, quantity has drastically out-performed quality when improving ML models. However, model training techniques are getting more interesting as projects like LookingGlass have been able to improve their generation on a now-outdated data set solely through experimenting with training models. I think it’s a case where a certain density of data was a prerequisite to show promising-enough results to get it started, and then data size has been so much easier to improve with that other fundamental research has been left behind.
One of the beneficial outcomes of people getting concerned about data scraping may be more investment in model training and understanding.
So what are the odds on one of these bot scraping the WIPs saved in Adobe Creative Cloud apps? Or scraping other web apps where people work on drafts of their docs that are not explicitly public?
anyone who has seen how social media algorithms and braindead SEO drivel have turned the internet into a violent and diseased wasteland and then thinks mass database driven AI tech will be a net positive for society is clinically insane.
you know lev manovich and friends were doing this kind of exploration at ucsd while I was a student there and it always seemed way more interesting when you knew the bounds of the dataset - they were doing this kind of analysis on like, van gogh’s whole life’s work and doing things like superimposing all the work on top of each other and filtering based on various parameters and it was a really interesting and unique way to look at an artists’ entire oeuvre without defaulting to the usual cliches that come up trying to describe their work purely aesthetically
maybe it says more about me that I find the same process applied to a dataset of The Internet super boring? like it only ever reaffirmed my thoughts about the internet and that’s not a great feeling!
ok I’m coming back to this because I think I’m thinking about this in a roundabout way - I guess what weirds me out about ai art is that we want the ai to do the exact thing that we would do if we were skilled artists, which is just to produce art - that just strikes me as the absolute least interesting use of the technology
maybe that’s what’s driving the bad feelings towards it - this sense of having a thing trained on huge datasets of all of the internet and using it to just draw pictures instead of literally anything else
If AI slop and mediocrity is the inevitable outcome it seems to me the correct response is to fight against it by surfacing and promoting those who resist it. There’s literally no way this process can produce any art worth a damn, and I think we need to start getting on the bandwagon of devaluing it now.
We need less content farms of mega corporations defining our world, we don’t need a better reason to reject this out of principle.
The crazy thing is that image generation is basically an accident. It doesn’t seem like something easy but it just kind of falls out of computer vision in a way that seems to have caught everyone by surprise.
I think there are a lot of undiscovered uses of this that exploit the way concepts are linked but in surprising ways. That Pokemon image generator plugged into a game (Game Emulation via Neural Network) I linked a bit back is a nonverbal demonstration of the underlying statistics: trained on Pokemon gameplay, once you hit a long forest path or ocean you land in a local minima because the most likely outcome of walking there is more of the same. It’s the responsibility of the chaotic targeter to get more and more agitated until it can bounce you out of it. And then, it’s really interesting in how it reminds me of my own less-consciously-directed dreams, getting stuck on imagery and patterns and changing bits of an image-space until the whole thing is unrecognizable. It’s the first computer-controlled space that truly feels free from the literalism imposed by code.
The best process I’ve seen comes from re-training the model, feeding output back into itself and digging a weird little tunnel
I was kind of wondering about the “chat” bots that are getting purposed for songwriting, would those need tags, and would those be scraping various docs sites?
I guess the weird thing to me is that there don’t even feel like any particular insights about automation or managerialism or machine art to be had by relitigating this stuff with whatever degree of enthusiasm or crankishness. Like we’ve all had those debates and they generally feel more interesting and alive in their original forms – maybe because they tended to originally be proposed by people rather than presented as drifting, inevitable secondary characteristics of the mystic alienated cultural shape known as Computer. We can have moral arguments about this stuff (about repurposed art etc) or social arguments (the impact of automation) or weird shadow debates about accessibility but it doesn’t feel like there’s very much to say about these things aesthetically – whatever cultural tendencies they’re wrapped up with, we never feel like we’re re-encountering those ideas in surprising and sinuous ways through an aesthetic encounter with the work. Which I feel like would ordinarily be something of a disqualifying statement in the context of art, right? And the fact that it’s not is kind of more interesting to me than the ins and outs of the programs themselves.
So I think what I ultimately find disconcerting isn’t really anything to do with “generated art” which has been around for the whole of my life and which people mostly never cared about. It’s the sense that this same stuff now suddenly has an automatic importance just as proxy for Tech, imagined less as a shadowy afterimage of alienated human potential than as something more active and real than ourselves, whose output is seen as so intrinsically significant that it all needs to be scrutinised for signs of the world to come in much the same way as all the other solutions-looking-for-problems like blockchain and nfts. the aesthetic lack to ai pieces doesn’t matter because the revelatory element of the aesthetic has sort of been foreclosed upon as something unique to tech itself, with whatever’s left just being a matter of inarguable and unimportant personal taste. I think the only thing it’s forced me to reconsider is the extent to which this manifests culturally as, Everything Is James Corden Now: somehow boring and degrading to hate while also being impossible to be enthused about; always prompting fantasies of some kind of nuanced or solving take which will absolve you from having to further consider him, while at the same time feeling aghast at the idea of expending your energies trying to come up with one; always present while somehow having no identifying qualities; ultimately just another load to be borne, with a kind of sullen peasant indifference towards whatever all-powerful yet inculpable system somehow made him our problem.