I’m convinced AI art is going to be a boon for artists. People always assume that the demand for something will remain static as its production grows and the producers become more productive. But as the cost of something falls, people consume more of it. That’s a law in economics. This is no different.
What’s likely to happen is people are going to invent whole new media that will use this. Incredibly elaborate interactive and personalized experiences will be made possible by this. One person will be able to deliver a huge amount of value, and they will be rewarded accordingly.
Think back to the 1980s. Programmers were in very short supply, and the work they were doing was less accessible and more difficult than it was today. IDEs, version control, better languages and frameworks, easy access to tools to learn to code, Stack Overflow, etc. made programming much more accessible and engineers became much more productive. The number of programmers out there is orders of magnitudes greater than three decades ago. Yet real salaries have still grown tremendously. The value per capita has grown even faster than those salaries because they’re so productive.
The same thing is going to happen here. Artists will build unimaginably huge, complex projects (on the scale of Google/Facebook/Amazon), and the world will gobble them up.
The lower the barriers to entry, the more competition game developers face, often by teams willing to charge little or nothing. Look at the app stores on mobile: 0.01% of apps make back their development cost
Steam had almost 9000 new games added to it last year, and the average revenue of each fell by almost half a couple of years ago
From an economic standpoint, perhaps. Consider the hobbyist.
If these tools had been available when I was a kid, my tile based RPG could have had the graphics I always wanted it to have. Instead, it never got anything more than map tiles (primitive ones at that). I had no talent or skills for asset creation.
Nobody played that game other than me, and (once) a couple other people I knew. That was fine, I made it for me, not others. But I would really have been happy to have access to assets.
Yes, creating art itself will have a lower barrier to entry, just like learning to code in Python has a lower barrier to entry than Assembly. But any software engineer will tell you that most of the challenges don’t come from writing lines of code, they come from building things that can scale at scales. Building a big thing in a big organization means having good communication and project management systems. Those things haven’t been taken over by AI (yet).
>But as the cost of something falls, people consume more of it. That’s a law in economics. This is no different.
I don't know such a law, but it's simply false.
People have 2 finite budgets they can allocate to entertainment, time and money.
In many ways money is not the limiting factor anymore since people have Steam lists with hundreds of unplayed games and the abundance of Free2Play. The main factor it not to have a saleable product, it is to manage to get in in front of enough eyes, with all marketplaces already being saturated and ads being useless unless you have a very high budget.
I'm a bit worried that it will devaluate indie productions even further because of an over-abundance of AI generated content, further elevating industry production that can afford really competent artists that you can't even start to replace.
I'd draw some parallels to the rise of advertising as well. For most people, the culture industry as entered at "post-scarcity" phase, and the consumption of mass-art is now gated more by eyeballs and less by actual production. The only (reasonable) way to increase eyeballs is through advertisement, paying 3rd parties to force consumers to see (part of) your art.
Yep. I have been gaming on native elf/linux (no msft grade proton plz) for a decade, and when we talked about time, it is a bit more complex: based on the type of game, it depends also on your state of mind (tired, motivated, etc). Games are "demanding", as it is not passively watching some content. I cannot play a game of dota2 if I am too tired (namely the adrenaline won't compensate enough), and you will need a break from it once in a while, if not burnt out for good.
Namely, it is more than time alone, it is "that right time" and there is much, much less of it.
I'm not sure economists have ever overeaten :) More is for sure not always better. For instance, there may be storage costs. I discourage people from giving me nonconsumable gifts, because I don't have enough space for the crap I already have. Other economic actors like corporations can totally have this problem.
"All things being equal" is doing the heavy lifting there. All things are no longer equal though, which is the point. The time/effort to produce art has effectively become zero.
> The time/effort to produce art has effectively become zero.
But not so the effort needed to produce "unique art".
When everybody can push a button to produce "art"
then how do you make the results of your button-pushing unique?
The value of art is in its uniqueness, and originality.
Using AI to generate art based on input training models is not that different from photography. Each generated art-piece is like a photograph taken from different viewpoint. And when everybody can use the same programs to generate their "art" the results are bound to be more or less non-unique, non-original.
The scary part for artists may be that now programmers can create new original generated works of art. You can be an artist without knowing how to draw.
Then again we could have a conventional artist with a unique style who would train an AI model to only generate versions of their existing paintings. Laws may need to be adapted to make it illegal to generate art based on someone else's existing art. Lawsuits will follow.
The effort required to produce 1 particular "bored ape" or any of their many many knock-offs tended to zero. They were technically unique, sure, both by url and by content. But also mass-produced, low-effort, cookie-cutter variations, cheap and ugly. As to the last, "ugly" - without getting into "But is it art?" - if it is; it's really bad art.
Ceteris Paribus (all things being equal) is one of the economist's favorite phrases, along with 'on the other hand'. And to be fair, it's hard to predict how a change will affect things if everything else changes as well. It is a huge assumption though.
i'ts not a law, but there is also Jevon's Paradox in which the more efficient use of a resource causes overall increase in the amount of overall consumption of the resource
Yeah, people always underestimate/forget the ability of artists to use everything for art in novel ways, I see no reason AI is any exception. I'm sure it's already happening and I just haven't been exposed to it yet.
>I’m convinced AI art is going to be a boon for artists.
Have you talked to any artists? Because in my experience they universally hate AI generated art. Not just a little. I think I've seen more artists who were open to NFT's at this point than AI. Art sharing websites are getting filled to the brim with posts from new accounts sharing AI art which sometimes very directly takes clearly identifiable features and styles from other artists.
AI generated art is beckoning a possible dark age for visual arts. There are many lost arts already, if AI trivializes the costs and efforts why bother honing it. Then we find ourselves in a creative void as the only art generated will be based on art that of which has come before.
Disclaimer, I’m a layperson when it comes to art. I’ve taken one art history class so I could probably talk about various art styles, but I’m no expert.
In my experience of visiting art museums, some of the classical and renaissance art is incredible. There is clearly some profound idea behind the piece, and the skill is the vehicle of that idea. A huge amount of art, though, is simply commissioned portraits of some wealthy businessman or forgettable Duke. The purpose of these paintings is simply to represent reality. There is no idea behind it.
For the artists painting them, the camera must have been devastating. It meant that years of training in studying how to properly draw an eye and how to get the colors inside a shadow right were at risk of being entirely devalued. But the camera also laid bare the fact that art is more than just skill. I imagine a lot of artists who were coasting by on skill alone hated the camera.
We’re at that point now with digital art. The cost of skill in actually drawing a thing is now zero. Anyone can do it. The question is, do they have anything to say?
Everyone has something to say and most of it is noise just like those portraits except now the gallery is going to be 1000 to 1 useless portraits to meaningful art. What is good will be lost in a sea of mundane as everyone tries for the spotlight.
Professional artist here, and I can tell you there are as many different opinions as there are people, including those who couldn't care less. But if you look at the hot take mic drop platforms, you will of course see either hype or outrage with little nuance.
It's happening very quickly and the outcomes are difficult to predict, but it's clear it's a big deal. We either fear the unknown or are excited by it. I feel most of the controversy is in fact the robots-will-make-our-life-comfortable / robots-will-take-our-jobs discussion. The real problem is elsewhere, not the tech itself.
Photography is if course a good comparison, but don't forget 3d CGI. I can place things exactly how I want them and have the computer do ALL the drawing for me? Outrageous!
It certainly removed a lot of need for menial manual work (and even filming), but has also opened the opportunity for creating entirely new things that before would just not be practical, or enjoyable to make. And there's no doubt now that a lot of skill is required to make things look good.
It can seem like it's easy to get great looking stuff with AI but it's not. What's easy is to churn out tons of samey derivative crap that will age very badly as soon as we get used to it, like with early 3d.
Yes, good point but let’s look at what a a boring mess CGI movies have become. Movie productions used to employ a lot of artists and looking back you actually can tell. For example the 1982 movie ‘The Thing’. All those effects in the movie are realistic, are not CGI and have become now a lost art.
CGI didn't crowdsource 3D models from other artists to dynamically build scenes with no technical effort and offer it to the masses for free. Doesn't seem comparable.
> There are many lost arts already, if AI trivializes the costs and efforts why bother honing it.
Photography can capture reality far better than any artist could achieve even with a lifetime of practice, and today it can do so with nearly zero cost--take out your phone, hold down the camera button, snap a picture. These advances hardly put an end to drawing and painting as skills.
It turned out that there's value in art beyond verisimilitude, and if anything the advent of photography encouraged a resurgence of exploration into what art really is: a movement away from pure verisimilitude in search of feeling.
This is what I expect to happen with AI art today. Yes, the AI models can produce art that looks awesome and wonderful now, but it's frankly already starting to wear thin. Art exists to fulfill humanity's need to experience something new, and AI alone simply cannot keep up with the meta game.
There will need to be human artists guiding the AI to produce something valuable, and there will continue to be human artists working by hand to create the novel works that the AI could never have rendered.
> There will need to be human artists guiding the AI to produce something valuable, and there will continue to be human artists working by hand to create the novel works that the AI could never have rendered.
Yes but who’d want their art to be fee back in trainig the AI to achieve what they have done with a click of the button?
> This is what I expect to happen with AI art today. Yes, the AI models can produce art that looks awesome and wonderful now, but it's frankly already starting to wear thin. Art exists to fulfill humanity's need to experience something new, and AI alone simply cannot keep up with the meta game.
But the essence of artwork can actually be copied away too and in the process squander the artists livelyhood. Yes, the argument that the artist can themselves take advantage of that and that is true for some computer literate artists but what about the others? Does the future of art necessarily have to tie art to computers and AI?
> But the essence of artwork can actually be copied away too and in the process squander the artists livelyhood.
No, because the essence of art is meaning, and AI cannot deliver that without a human agency behind it.
AI will do to art what technology does to every field: it will eliminate the need for shovelware-style jobs, leaving only the truly creative jobs behind.
It will most likely resemble what happened to web developers when SquareSpace and WordPress became good enough for untrained people to pick up and build with. If all you knew how to do was write enough HTML and CSS to make a website that looked "good enough", your job was in danger. But if you were a good designer or a good engineer, you transitioned to a more specialized role, working for the many orgs who needed something beyond the capabilities of the tech.
I expect the same will happen in art. To use book covers as an example: the job of illustrating covers for cheap romance novels will certainly be replaced, because they all look the same already. But publishers will still need humans (either driving the AI or painting by hand) for the books they hope will be best sellers, because it takes a real artist to know how to capture the spirit of a book in a cover image.
>No, because the essence of art is meaning, and AI cannot deliver that without a human agency behind it.
I see no reason why another AI eventually couldn't come up with prompts from human interests found from algorithms. There's exactly no reason a human needs to be involved at all.
I am an (amateur, non professional) artist and I am a big supporter of AI generated art tools.
The main issue is that anyone can write a simple prompt and get 85% of the way to an image that previously required a decade of experience to create.
But that last 15% is where real artists differentiate themselves. It’s the last 15% that can’t be faked if you don’t have a sense of artistic direction and discretion.
I strongly disagree with “why bother honing it” as well. If anything, being able to get to the “honing” stage so quickly is what makes these tools so powerful. I’m sure if Gordon Ramsey had to butcher his own chicken, grind his own grain for flour, and churn his cream into butter, he wouldn’t spend as much time on the finishing and plating of a chicken sandwich.
I have talked to and worked with a few artists, trying to capture their style using generative AI models.
With the limited subset of artists I have worked with (traditional modern / contemporary painters) they have been very inspired by the creative possibilites of these models and in one case see it as a great opportunity to scale their work to new types of media such as video.
It would have been practically impossible for one of the artists I am working with to draw each frame by hand but now she can generate music videos in her style by tweaking and working with Stable Diffusion and a set of reference images.
Have you looked at the resulting art works? There's an exhibit called "Unsupervised" currently running at the New York Museum of Modern Art [0]. I saw it last week--a fascinating of collage of shifting, waterlike images with delightful trompe l'oeil effects. If this is what AI-generated art can look like I'm all for it.
>AI generated art is beckoning a possible dark age for visual arts.
I don't think the invention of a camera displaced many artists, but I'm quite sure that they were just as angry, probably saying how it doesn't have a soul and such. But in the end it improved the life of humans to an unimaginable levels.
I think same if not more can be or will be achieved with AI generated art. And I don't think it's displace many artists jobs.
Again, the coming of AI art feels like the coming of camera must have felt.
More specifically, the invention of print processes suitable for reproducing photographs displaced a huge number of artists. If you go looking through glossy magazines of the 1940s, they were packed with illustrations.
The invention of the Camera was drastically different. Realistic portraits and actual photographs are just not the same thing and have their own respective places.
AI threatens to make artists obsolete by via the direct usage and synthesis of their work along with the work of their peers. Used without their permission.
I'm curious how many people have actually tried these networks out. I've dropped a bunch of demos below so you can actually try them and I encourage everyone to do so before they form strong opinions here.
As someone that works in generative modeling I think it is important to note that we are hand selecting results and not showing the average ones. There's been such a hype machine that if we show our average results that we'll get killed in review (publishing in a hyped space unfortunately requires a lot of salesmanship). Not a thing I'm particularly happy about, but I think this is important to note because if you're just looking at the images on Twitter, Blogs, and other social media then you're seeing the top 1% results. Images that take quite a bit of time and effort to produce. Likely not as long as an actual artist to make them, but still a lot of skill is involved.
I do agree that I think it'll be a big boon for artists. I've even been using these to get better at drawing. The results are bad even with prompt engineering, but hey, my human mind can fill in the gaps and it is good for expanding creativity. We'll get better at these models for sure but I'm not quite convinced it'll kill a lot of art jobs. But maybe I don't know what those jobs are that are being killed. But then again, I've even seen artists do much better at these models than me even though I understand the code and math going on. There's also cool things like walking through the latent spaces, but you won't be able to do that with the hugging face code. But that is art that you won't be able to do elsewhere. It is entirely new.
I tried Ghibli diffusion to get some decent cartoon styled results with SD, but it's really bad. Results were all hallucinations from the training data, no interesting new combinations, no novel prompts that would work. I stopped after a few dozens.
I'd still like to get some good cartoony results with SD, but it's clearly not possible yet.
It is still possible. I've gotten some really good quality stuff. The thing is you still need a lot of work and you're often going to have to play with parameters hugging face may not give you access to. We also generally generate images in batches and select the better ones lol. But that's why you see DALLE give you several images of the same prompt.
I've been more messing with DALLE and in those specifying style is really useful. For SD I think the best thing is to follow some of the Twitter accounts that are posting prompts with their images (though they aren't posting other settings usually). Twitter on my computer is not working right now so I'll try to remember to come back later.
But suggestions I'll give is that you need to add detail to the prompts. SD also has negative prompts which really help. So use a lot of words if you can (twitter accounts will help you find magic phrases like "bismuth neon chrome copper patina verdigris"[0]). But you may even see here that the magic words with SD aren't as simple as with DALLE
I do think you're understanding, though, what my post is about and why I think a lot of people are talking from a naive perspective. If you only show the wins you have some extreme biases.
Interesting magic word. My prompts were often full of 'masterpiece' or 'octane render' (which improves 3D consistency).
With regard to your main point, I generated more than a thousand images trough various models, and yes good results such as these posted on social media are very rare. And people definitely spend lots of time to adjust things with inpainting and other tricks, in a way that I would find infuriating. For the effort it takes I would rather just redraw it. It depends on the styles; with brushstrokes you can hide a lot, with cartoon lineart is crucial so fixing requires more work.
What’s likely to happen is people are going to invent whole new media that will use this. Incredibly elaborate interactive and personalized experiences will be made possible by this. One person will be able to deliver a huge amount of value, and they will be rewarded accordingly.
Think back to the 1980s. Programmers were in very short supply, and the work they were doing was less accessible and more difficult than it was today. IDEs, version control, better languages and frameworks, easy access to tools to learn to code, Stack Overflow, etc. made programming much more accessible and engineers became much more productive. The number of programmers out there is orders of magnitudes greater than three decades ago. Yet real salaries have still grown tremendously. The value per capita has grown even faster than those salaries because they’re so productive.
The same thing is going to happen here. Artists will build unimaginably huge, complex projects (on the scale of Google/Facebook/Amazon), and the world will gobble them up.