I've been using Stable Diffusion heavily for a couple months now, and I'm getting increasingly frustrated by the backlash and smear campaigns from people who are ill-informed about what this...
I've been using Stable Diffusion heavily for a couple months now, and I'm getting increasingly frustrated by the backlash and smear campaigns from people who are ill-informed about what this technology is and how it works.
This is pattern recognition software. When you train a model, the software ingests images and metadata about them. It looks for commonalities between annotations (image captions/descriptions) and between the images themselves. It assigns probability weights to those commonalities, meaning it comes to associate certain graphical characteristics with certain words or phrases (tokens). It's able to find more and richer patterns if the data set it's trained on is large and well curated.
This is a mathematical distillation of the same process human artists go through as they develop their skills and personal styles. Nobody picks up a paintbrush and creates a masterpiece on the first try. Great artists only become great by exposure to and study of the works of other masters that came before them. You don't grok the finer points of shading, perspective, composition, style, etc., without observing and comparing what others have done first.
This technology does not save images in a database and composite them together into collages. The Stable Diffusion model is trained on upwards of 100TB of images, but the model itself weighs in at a tidy 4GB. It's literally incapable of clearly reproducing any of the images it was trained on. What it can do is synthesize new images by starting with a field of random noise and iteratively modifying it until patterns emerge which correspond to the weights it was trained on, with its attention during this process probabilistically steered by user-supplied text prompts.
This is, again, a mathematical distillation of human inspiration. We learn about the world through observation, and the act of creating art is to transform a blank canvas into something unique that meaningfully represents the understanding we gained through those observations. If this is plagiarism, all artists plagiarize; everything is a remix. I can empathize with an artist wanting to withhold her works from a training set. But to me that's like a painter showing their gallery to another painter while saying "you may look at my paintings, but may not learn anything from them, lest they influence you into painting something of your own later on that I deem too similar to mine."
Now, after saying all that... this is a tool, and an imperfect one. Despite claims that "you just push a button and get a masterpiece," working with SD is an exercise in patience. It's laughably bad at understanding details like multiple subjects, relative positioning, scale, composition, and anything intricate including hands and feet, not to mention the barest understanding of basically any designed object. The flaws in the training data and annotations are readily apparent, it has blind spots a mile wide and produces freakish garbage more often than not. Any output with text in it is completely unusable. It defaults to 512x512 resolution gens and compute resources climb astronomically if you want more than that.
My workflow involves painstakingly crafting a prompt that's both detailed enough to produce something interesting, and general enough to sidestep the gremlins that lurk inside SD when you overspecify. I produce countless low-res gens as I go, checking and correcting my assumptions. Once I'm reasonably happy with a prompt I use it to generate 100-200 high-res images and then sort through those manually to find anything good enough to work with. Then I'll pore over whatever's left, manually cleaning up any remaining AI weirdness, janky hands, nonsensical backgrounds, oversmooth textures, bad cropping, and so on. I use a combination of AI inpainting, outpainting, and good old fashioned Photoshop. This is art. It's far from effortless.
And here's where I think artists need to get on board. These are incredibly powerful new tools for your toolbox. Just as the art world reacted to and then embraced digital bitmap and vector formats, and the innumerable analog innovations that came before them, so we need to recognize what this is and avoid throwing out the baby with the bathwater. In the hands of a skilled craftsperson, an AI toolset is a game-changer. I'd encourage all artists to play with the unified canvas in InvokeAI and see what they can do with it. It doesn't even need to involve new image synthesis, it's astonishingly good at manipulating existing images, for example restoring damaged photos or quickly blending in manual edits.
The fact is, Pandora's box is open now. You can't uninvent this technology and it will only get better over time. So do we dig our heels in and resist change? Or adapt and allow our own horizons to expand with it?
It took the subject of this article three days to make a children's book and start selling it on Amazon. Do you know the normal timeline for this sort of thing, making art, writing the text,...
It took the subject of this article three days to make a children's book and start selling it on Amazon. Do you know the normal timeline for this sort of thing, making art, writing the text, coordinating between artists and publishers? A year, minimum. The people whose lives are going to be upended by this are angry because they're scared, and they're scared because this will completely change the industry and their lives. They don't care about any of the technical things you mentioned, just that the skill they have spent a lifetime developing (often through significant personal hardship, given how little most artists and writers made even before this) has been suddenly devalued. Why would anyone buying art care that there are flaws that need a little cleanup or that there's some work to choose between a hundred best attempts when Stable Diffusion and ChatGPT will take you 95% of the way there? Their lives will never be the same.
I will push back on the idea that artists will or should get on board with these trends. These programs are great for putting out something that matches imagination to existing artistic or narrative styles. But (at least for now) if you have a vision that doesn't match what's come before you've got to have a person involved. How will those people make a living creating art long enough to develop their own personal style? Where will new themes and visuals come from to feed the gristmill of automation? It's all well and good to automate art, but these programs may be the artistic equivalent of clearcutting a forest; if you harvest everything now, you'll stop new growth.
I need to recognize that "AI art will devalue existing artists and threaten their livelihoods" is a separate argument from "AI art is plagiarizing existing art." I was mainly intending to make a...
I need to recognize that "AI art will devalue existing artists and threaten their livelihoods" is a separate argument from "AI art is plagiarizing existing art." I was mainly intending to make a case against the latter. It seems that a lot of discussions about this online are conflating the two, which doesn't do anybody any favors.
I agree with the former point, actually. AI art is a threat to the art world as we know it today. The footnotes of history are filled with epitaphs of industries killed by innovation. The automobile decimated the livelihoods of farriers. The electronic computer made human computers obsolete. How many telegraphists were put out of work by telephones and fax machines? I don't intend to diminish any of those professions. These were hard-working, deserved people with specialized knowledge, who had the rug pulled out from under them by circumstances beyond their control. No one's hardship should be celebrated.
Still, somehow, we soldier on. Those who can read the writing on the wall are wise to adapt. They can carve out new niches for themselves. In many cases, instead of being crushed by a new technology, there are those who find a way to springboard off it to greater success. On the other hand, those who indignantly try to keep on as they always have will go down with the sinking ship. Again, I don't mean this with any ill will. I only mean that, in terms of career survival strategies, life sometimes imposes big choices on us, whether we want to face them or not.
As a side note, I've really enjoyed Jason Feifer's Build For Tomorrow podcast. Each episode pulls an example from history of a disruptive change, like the invention of the bicycle, and examines the naysayer response at the time. Spoiler: The same arguments are made every time. There's nothing new under the sun. This is at least an easier pill to swallow when viewed in the context of history instead of the tenor of in medias res flamewars online.
Self publishing, as the subject did, takes much less time. A friend of mine is an author for a living, he turns out 3-4 novels a year with cover art, solo writing and editing with an artist for...
It took the subject of this article three days to make a children's book and start selling it on Amazon. Do you know the normal timeline for this sort of thing, making art, writing the text, coordinating between artists and publishers? A year, minimum.
Self publishing, as the subject did, takes much less time. A friend of mine is an author for a living, he turns out 3-4 novels a year with cover art, solo writing and editing with an artist for the covers. I doubt the publisher is using up 9 months to get their end done, so if someone can't manage a children's book once a quarter then they're really freaking slow.
The people whose lives are going to be upended by this are angry because they're scared, and they're scared because this will completely change the industry and their lives. They don't care about any of the technical things you mentioned, just that the skill they have spent a lifetime developing (often through significant personal hardship, given how little most artists and writers made even before this) has been suddenly devalued. Their lives will never be the same.
Should we lament the passing of town criers, scribes, carriage makers, blacksmiths, or the most obvious in regards to art and technological advancement: portraitists - almost entirely replaced by photographers (who have their own arguments to make on whether or not they are artists).
At one time art required physical media, paint, ink, chalk, and the like along with the training and skills needed to wield those tools. The rise of digital art, illustrator programs, and photoshop hasn't stymied art, it's given art the accessibility that nothing before it ever could have. How is AI generated art any different? It's a tool used to provide an output of the input given. A paintbrush outputs a slice of color based on the inputs of the hand that holds it.
And as with all things, there is ethical consumption. I choose not to eat factory farmed meat, caged eggs, or buy slave labor electronics. People that value art will do the same and choose not to purchase AI art and support those that share those values.
I will push back on the idea that artists will or should get on board with these trends. These programs are great for putting out something that matches imagination to existing artistic or narrative styles. But (at least for now) if you have a vision that doesn't match what's come before you've got to have a person involved. How will those people make a living creating art long enough to develop their own personal style? Where will new themes and visuals come from to feed the gristmill of automation? It's all well and good to automate art, but these programs may be the artistic equivalent of clearcutting a forest; if you harvest everything now, you'll stop new growth.
Since all art is derivative, humans artists are nothing more than biological AI, just as incapable as the AI of creating something truly new. We've just given names to the derivative styles that can all be eventually traced back to our origins. If prehistoric humans didn't slap their hand on a cave and spit dye at it we wouldn't have Monet. AI doesn't change the fact that to produce something truly new requires a human. If anything AI is giving humans access to more artists than ever before since it was trained on more artists/art than a single human could ever experience in a lifetime. Putting in a prompt has the potential to expose the prompter to more art styles than ever possible prior to now, it's a tool. One that can be used to inspire or create.
As for how an artist should make a living until they develop their own style (of the very few that ever do), they do so the same way anyone else does, by doing lower paying jobs until they have the skills necessary to demand high paying jobs. Actors, comedians, singers, etc. all have to work up to the positions they want doing things that are, more often than not, unrelated to their chosen profession. Why should artists or writers be any different? But in the same breath, artists especially have the advantage of patrons. No one (hyperbole) is giving Suzy the waitress a patreon subscription until she gets her big break, but there are thousands upon thousands of artists with such subscribers, supporting their lives wholly or until they can survive on the income of their wares alone. I'm one of those people that is a patron of artists doing so because I believe in their work. A friend of a friend is a real life patron of the arts, in the centuries-past definition of the word, of an opera singer; paying to supplement her life while she trains, travels, and performs.
Is there enough to go around for every person that wants to be an artist to live off of their art alone? Debatable, but unlikely in this world of scarcity.
Funnily enough for your analogy, forests require clearing from time to time for new growth to occur. If trees do not fall or fire ravage the valley then sunlight does not reach past the canopy and nothing new can flourish. Deviantart alone has 61million active users, even if we assume only 10% are actually creating anything, that's still 6 million artists churning "tens of thousands of original pieces of art every day". How much of it is even worth the hard drive space it takes up? Just because something is made doesn't mean it's worth anything. Perhaps AI is the forest fire needed to clear the canopy and let the sunlight through.
Furthermore, humans are affected by their sociopolitical contexts and anxieties: war, racism, sexism, poverty, ennui. And human creativity also refracts through human biological impulses:...
Furthermore, humans are affected by their sociopolitical contexts and anxieties: war, racism, sexism, poverty, ennui.
And human creativity also refracts through human biological impulses: maternity and paternity, sexuality!, jealousy, and whatnot.
Can you define those terms further and maybe provide some rules that would let us test if a system is imaginative or not? If we were looking at a sentient and imaginative system, how would we know?
Can you define those terms further and maybe provide some rules that would let us test if a system is imaginative or not?
If we were looking at a sentient and imaginative system, how would we know?
The art world will get upended. But this isn't the end of new styles. At some point people will develop new styles by slowly creating a dozen images in the style and then fine-tuning a model on...
The art world will get upended. But this isn't the end of new styles. At some point people will develop new styles by slowly creating a dozen images in the style and then fine-tuning a model on that style.
Technology has always revolutionized art. Photography pushed artists from realism to Impressionism and Modern Art. Modern phone based cameras are revolutionizing photography. Along the way artists...
if you harvest everything now, you'll stop new growth.
Technology has always revolutionized art. Photography pushed artists from realism to Impressionism and Modern Art.
Modern phone based cameras are revolutionizing photography.
Along the way artists who were invested in old ways of doing things struggled, and artists who adapted and changed thrived.
Yet there is still a room even today for photo realistic drawings, wedding photographers, portrait photographers, Photoshop experts etc...
I've been thinking about the role of art in society quite a lot over the years. While I'm at least in theory agreeing with you, I do have some observations that I would like to share. When it...
I've been thinking about the role of art in society quite a lot over the years. While I'm at least in theory agreeing with you, I do have some observations that I would like to share.
When it comes to art of any sort, the actual value is entirely intrinsic. Which is to say that it doesn't actually have any monetary value. There are many types of art in the world but most people don't have the ability to appreciate the majority of them. For instance, sewing is an art, but most people look at clothes or other sewn fabric goods as art pieces; they don't appreciate the worksmanship and creative choices made when deciding which stitches to use.
One of the things that has historically been used to give value to any given piece of art is the amount of labor it took to produce it. And for a great deal of graphic arts, that's one of the most common ways that art gets valued; an artist receives a request to make something and then they consider their materials and time and give the requestor a price based on those. We've seen countless ways that this kind of arrangement has been broken, though. The development of photography killed off the business of classical portraiture because now there was a device that can capture lifelike images in seconds instead of weeks. Lace has almost entirely disappeared from fashion because it has transformed from a difficult time-consuming artform into something a machine can complete in a few minutes.
But at the same time we know that value can be subjective. That's why (in theory, if it's not just a money laundering scheme) there are pieces of art that be worth obscene amounts of money. And I think that understanding that is how artists are going to be able to drag themselves out of this situation. There is one thing that human artists have that AI does not, and that's intentionality.
I'm sure we all know that these types of AI are imperfect and they will probably get better over time, but they lack fundamental things that humans do. There's no dataset that will cover the entirety of a human's experience. They're also really bad at understanding what it is that you actually want; if you go even the slightest bit into the realm of imagination it will come up with strange deformed outputs. They're also really poor with accuracy; if you tell it that you want something in a specific space it won't understand you. Just for fun try asking it something you would expect a computer to be really good at drawing; try "a single black pixel" or give it an equation for a line. In a matter of speaking, the algorithms take inspiration from your prompt, but then they draw whatever it is that they feel like.
I expect we will always appreciate human performances, but in different ways than we do now. As technology advances, we have more choices about when and how we appreciate performances. We have...
I expect we will always appreciate human performances, but in different ways than we do now. As technology advances, we have more choices about when and how we appreciate performances. We have more options and it becomes increasingly optional.
Historically, this happened via the invention of printing, the ability to record and replicate music, and the rise of the film industry and TV. A superstar can become a billionaire but the average artist doesn't earn much. You no longer need as many musicians to let everyone listen to as much music as they like, while charging nearly nothing.
On the other hand, the ability for millions to enjoy the same movie means that more money can be spent on them, so there are more specialists. All the people you see in movie credits get a job of some sort. (Or if you want lots of dancers, Bollywood has you covered.)
It can go the other way too. Big bands are expensive so you tend to see smaller bands these days, or just a DJ.
So there's a question about which human performances are the ones people care about and will pay for, and which ones are superfluous? Also, are you getting what you paid for? What counts as "cheating" depends on what people agree to as the rules of the game.
I wouldn't expect all artists to shun these tools. Some will surely defect, invent new art forms with new rules, and find an audience that appreciates them.
I hate that internet discourse descends into death threats and abuse so easily. It's not even productive or effective compared to just talking things out reasonably, like what? Anyway, my first...
I hate that internet discourse descends into death threats and abuse so easily. It's not even productive or effective compared to just talking things out reasonably, like what?
Anyway, my first thought reading the discussion here is the contrast between AI art opinions here on tildes (demographics lean more tech) and on tumblr (demographics lean more creative) is so stark.
To be honest, I can agree with arguments about how the march of technological progress may come at the cost of livelihoods and of course, how capitalism negatively plays into this. But I don't really see any comments (yet?) on how this progress may come at the cost to culture if AI developers and regulators fail to protect artists' roles and relevance.
Artists' main concern may be their livelihoods but they also suffer through the carpal tunnel, demanding deadlines and clients, lack of opportunities and employment benefits like insurance... because they still love what they do. They enjoy both the process of creating and because their labour gives rise to beautiful, unique and impactful products that they can feel satisfied with, especially with the potential social audience and financial rewards. Keeping in mind even pre-AI, many budding and professional artists wouldn't reach half the audience or compensation their work deserves.
As a writer, any story I write could be reductively described as a "remix" of stories, tropes and culture that have existed for centuries. So I have to believe that my take on it, my version, my individuality, my perspective and my vision, will be just unique and different enough to be meaningful and valuable to some one else out there. At the heart of art is the human drive to make something they thought of and show it to some one who'll care about it. (Which will probably and hopefully remain possible even in a world with AI dominating the markets.)
But regardless, I feel like flooding the world with AI art (and writing) devalues that creative passion significantly in an already difficult and oversaturated market. Competing with AI turns the creative journey into an inefficiency, and art into a commodity. The role of the artist is reduced to that of a hobbyist or at best, a small-scale artisan, with limited visibility beyond that. The individual cultural origins and impact of an artist's work no longer matters to an average buyer or studio. Most artists not good enough to survive, would have to quit. Get a different job, maybe one in tech.
Because no matter how hard they work, an AI can do something similar, faster and cheaper. And when it's similar enough to be "good enough", the number of people who care that it's not traditional human-crafted art will be fewer. Traditional art itself will be harder to find, harder to see... harder to learn and harder to teach.
So, yea, I'm on the side of those scared to lose the jobs they love, because I don't want a world where we have fewer artists. Where the bar to succeed is set impossibly high and only certain artists with enough mass appeal will be able to keep creating for a living. Less originality and less diversity and less authenticity will only lead to a more and more uniform cultural landscape.
It makes me think of how we've seen CGI effects and animation start to dominate Hollywood indirectly because 2D animators and practical SFX designers are unionised. It makes me think of Disney's oligarchy on blockbuster movies that never challenge the status quo too much and how nothing seems fresh or original in cinemas anymore and yet mediocre movies continue to make billions. Technologically, they're excelling but culturally, how are they doing?
But maybe I'm wrong, maybe regulations will protect artists and AI will be an amazing tool of creation to strengthen artists' rights and working conditions, and not a weapon for more corporate exploitation. Maybe everything will turn out mostly okay and artists will be just fine. I'd love to be wrong!
Ahh, I'm glad! That's the main reason I find myself posting on Tildes, tbh, because I know there's a limited demographic active here but I like the culture and while I don't usually have anything...
Ahh, I'm glad! That's the main reason I find myself posting on Tildes, tbh, because I know there's a limited demographic active here but I like the culture and while I don't usually have anything to say, I hope my offering other perspectives is welcome.
And in fairness, I myself lack the full expertise to know how AI could be used in different ways, and reading this thread did enlighten me to the potential it has to elevate art in the hands of creatives and I truly think if not for the overwhelming context of capitalism (specifically in tech and in dealing with art & copyright & digital rights), there could be so much more good to come out of this.
Human generated art is too. No one produces an original work from nothing. The difference here is the ease of access. AI tools have opened art production up to everyone, not just those who spent...
AI-generated arts is still at its core derivative of human-original works
Human generated art is too. No one produces an original work from nothing. The difference here is the ease of access. AI tools have opened art production up to everyone, not just those who spent years training (learning to derive existing work). It's understandable people with huge existing investment are threatened by the devaluing of that work.
Studios are going to lose their monopoly on production, and the influence the guilds have on blacklisting people who don't go along with their collective. That will probably generalize to all...
Studios are going to lose their monopoly on production, and the influence the guilds have on blacklisting people who don't go along with their collective. That will probably generalize to all forms of entertainment, news, and propaganda too. Check out the AIs that generate 3D models and it's pretty clear that a photograph is all one will ever need to create an entire movie set or video game level in just a couple of years, and it'll take mere seconds to do either one.
Check out the list of AIs in the top comment on that video that can provide everything one needs to produce a video game. Now mix in the insane capabilities of ChatGPT and it won't be long before the AIs can do the coding too. That bot is already helping people with their PhD thesis and getting it right.
It seems democratizing, at least to me. I can sit here and create anything I desire without interference from another human being becoming necessary. I don't have to give up a single one of my rights or freedoms or sell out any of my principles to get investor money, or change things to placate some group of gatekeepers. I can sit down and tinker, and in a couple of months I've got an Avatar movie series or a Warcraft MMO or a new DC comic book series.
At that point, just sorting through and curating a million movies, a hundred thousand games, a billion paintings, and tens of millions of books a month becomes our new problem. What do you do when more masterpieces come out in a single month than a human can experience in their entire lifetime? I guess you get an AI to engineer us a longer lifetime. :P
The political implications of this are mindboggling, as it will become increasingly more difficult for any group, institution, or government to exert overt or covert influence on creatives as time passes. No gatekeeper can control that many gates opening that fast all over the world. Our grandchildren will probably have the same creative options as any vertical monopolist entertainment company has today, all from the comfort of their favorite chair in the basement - as long as they have the leather patience to guide the machines.
At that point, I guess it comes down to a simple question - do you have anything worth saying?
Tech is starting to speed up again - but this time, it's not just the hardware. AI is moving faster than ever and we're starting to see the various generations of any given narrow AI one-up themselves on the order of months, not years. Oh, and the need for large training sets is not necessarily going to remain a requirement for long. Our silicon is going to hold us back the most, but eventually that will give way to graphene, silicene, and photonics.
That new hardware will probably be designed by an AI as well, and it won't take the 60 years it took humans to iterate from the IBM 1401 to Threadrippers. It'll do it in twenty, perhaps less, and it's likely that newer hardware will also make inroads on quantum computing. Photonics is a natural fit for that task, as it satisfies more of DiVincenzo's criteria.
Kurzweil's prediction from the early 1990s of 2029 as a ballpark date for an AGI passing the Turing test and 2045 for a singularity-type event is starting to look dead on target now, not ridiculous. Capitalism has no brakes, either. The next two decades will likely be the most interesting and terrifying decades this planet has ever seen.
I'm happy for tech to be democratising! And I really appreciate your comment explaining how it could be so, I'm not familiar with the overall AI vision and the detailed projections and...
I'm happy for tech to be democratising! And I really appreciate your comment explaining how it could be so, I'm not familiar with the overall AI vision and the detailed projections and possibilities.
But I also see the reality of what tech has and hasn't done for democracy where we can barely control misinformation and hate (and that's if the guys in charge actively want to commit to it, here's looking at Twitter and Facebook) and regulations are too slow to enforce accountability so tech responds to markets primarily.
I don't have anything else to add myself, I think your points are just as valid, it's a complex topic with a lot of grey nuance and ultimately, we'll each speculate on the future with as much optimism or pessimism as we individually feel inclined.
I find it hard to believe that someone can work in such a relatively small industry and not be aware of how much crap there is in it. While there are a huge amount of amazing kids books out there,...
she is “concerned that the use of AI in creating stories will create a proliferation of poor-quality stories, both on the writing and the illustration side.”
I find it hard to believe that someone can work in such a relatively small industry and not be aware of how much crap there is in it.
While there are a huge amount of amazing kids books out there, there are SO MANY terrible ones. At the worst end there are entire publishing houses which seem to exist purely to turn out cheap, rubbish cookie cutter crap. Those jobs are 100% going to go to people prompting ML systems, but they were worthless jobs anyway and it will probably make the books slightly better. But ChatGPT etc is not going to replace Oliver Jeffers or Lauren Child, Jon Klassen or Shaun Tan. ML still can't have ideas and this is the thing I think so many detractors don't seem to realise.
“I don’t feel he deserves to earn any money from the book, because he has not actually put much work into it,”
I don't entirely see how long something takes to make, or how much effort it involves should make any difference. Terry Pratchett used to turn out a couple of full length novels a year, didn't make them any less good. Meanwhile Harper Lee wrote two books over 55 years, one of which was a bit crap.
If it really does take three days to make a passable children's book that makes money, then it seems likely that there would be competition to churn out more books. (Much like happened with app...
If it really does take three days to make a passable children's book that makes money, then it seems likely that there would be competition to churn out more books. (Much like happened with app stores.)
More likely, they won't make money, other than a few hits that go viral. And this is pretty much the case already for self-publishing.
A fifteenth century technopanic about the horrors of the Printing Press.
It was okay that the act of copying was hard. It built character, in Trithemius opinion, the same way as chopping wood (though to this “interior exercise,” i.e. exercise of the spirit, he assigned far more importance). For monks, labor was part and parcel of devotion, and if you weren’t good at writing, you could do binding, or painting, or for heaven’s sake practice. And it goes even further: the labor of manuscript writing was something for monks to do — for there was no greater danger for the devout soul than idleness.
For among all the manual exercises, none is so seemly to monks as devotion to the writing of sacred texts.
And people said the same thing about computers assisting in design and illustration. Back in his day, my ~65 year old graphic design professor had to paint letters on a can by hand for one of his...
And people said the same thing about computers assisting in design and illustration. Back in his day, my ~65 year old graphic design professor had to paint letters on a can by hand for one of his projects. But there is a difference. When computers started getting involved, he was excited, just imagine how much more a designer or artist can do when they're not wading through the tedium and difficulty of executing tiny, fiddly details. But they're still being wholly created by a person who has to understand all of the same art fundamentals as every person before them regardless of the medium or tools they had available. Whether you're putting chisel to stone, brush to canvas, or stylus to digital tablet, there's a vision you're bringing into existence that requires hard-earned skill.
That's just something you can't do by typing a few words into a box. Creatively, it really is functionally equivalent to looking something up on Google Images. Maybe you can find something like what you had in mind, or maybe even something close to what you had in mind, maybe something that inspires you in a new direction, maybe you even find something you straight up like better than what you had in mind. But you can't use this type of AI to create something of your own, you can only sift and hope to stumble on something good enough.
They have their uses, for example one of my graphic design projects was pushed in a new direction by something I saw in a smeary AI soup. Another project was greatly helped when I was struggling how to even represent dirt in an Art Noueveau-like style. I couldn't find any references and everything I was doing looked awful, but a few rounds of harassing an AI gave me something I thought I could work with. But at least right now they're not the next technological advancement in creative tools. Because you really can't use these to create something that's yours. It's just an endless Google Images search on shrooms.
I have a lot of thoughts about some of this stuff. Forgive me if I ramble... There are lots of kinds of art. Some folks really like Jackson Pollock or other types of non-representational modern...
I have a lot of thoughts about some of this stuff. Forgive me if I ramble...
There are lots of kinds of art. Some folks really like Jackson Pollock or other types of non-representational modern art. In fact my sister-in-law is very into acrylic pours and other types of abstract art. I would argue that art pieces like Pollack's, or acrylic pours, are as much created by basic physical processes and random chance as they are by a person's skill. You have to know how to set up the process, and there's a certain amount of experience that you can leverage to make sure the outcome is more pleasing to the eye, but at some point gravity and fluid dynamics are doing as much or more of the creating as the artist. Yet I don't really see anybody arguing that those pieces aren't art just because most of the actual composition is done by gravity.
Andy Warhol famously created art pieces by remixing existing photographs and silkscreening them or otherwise remixing them. Does anybody seriously contend that Warhol's pieces aren't art? (They may, I genuinely don't know. I suspect not, though)
And none of those things invalidate other art forms that require more human input or human skill. Traditional oil painting still exists. Sculpture still exists (alongside 3d printing). Scrimshaw and weaving and hand sewn fabric arts and any number of primitive, skill and labor intensive arts and crafts still exist even though we have made it very simple to create most of those things with machinery and computers.
I do think there is danger than ML and automation are going to cost people jobs. In fact I think that automation is going to be a threat to a lot more than just professional artists' livelihoods in the coming decades. We're going to need fewer lawyers and paralegals to create boilerplate contracts. There will be less demand for copy writers, programmers, cashiers, call center workers, and God only knows who else. Sure, automation and machine learning will create a few new jobs, but not as many as it causes to be lost.
I think we're in for upheaval for a lot more than just artists and designers. If we react correctly to what's coming, if we create social safety nets and redistribute the wealth that is created by automation, then I think people will need to work less and will have the time and the peace of mind to create things like art, music, and writing just for their own sake and not so that they can pay rent and buy food. I'm not optimistic, though, that our political systems are capable of working toward that kind of collective long term good. I hope I'm wrong about that.
I don't think anyone doubts that this will cost jobs. However, it is possible that, in the future, AI art will be considered an overall improvement and a fundamental aspect of our lives.
I don't think anyone doubts that this will cost jobs. However, it is possible that, in the future, AI art will be considered an overall improvement and a fundamental aspect of our lives.
I actually agree with you, at least on days when I'm feeling more optimistic. And full disclosure, I love AI image generators, I've been in betas for several of the models that are out right now...
I actually agree with you, at least on days when I'm feeling more optimistic. And full disclosure, I love AI image generators, I've been in betas for several of the models that are out right now and I spend a good bit of time playing with Stable Diffusion and various models that have been built on it. I think that ML and automation could bring a lot of positive things in the future, but I do worry that there are going to be some pretty bad growing pains. I'd love it if we could be forward looking enough as a society to see how mass automation could and probably will affect us and put safety nets in place for the folks who will be the most vulnerable, but I'm not always optimistic that we will be.
Everything will eventually settle, but I agree with you that it is unfortunate that some will have to pay more than others. There should be mechanisms in place to accommodate those harmed by such...
Everything will eventually settle, but I agree with you that it is unfortunate that some will have to pay more than others. There should be mechanisms in place to accommodate those harmed by such changes. Universal basic income would help.
Commercially speaking, there was a recent copyright dispute between the photographer, and Andy Warhol's estate. A judge found that Andy Warhol's screen prints were transformative enough to not be...
Andy Warhol
Commercially speaking, there was a recent copyright dispute between the photographer, and Andy Warhol's estate.
Raises hand. Most of, and certainly almost all of his most famous pieces, are the art equivalent of selfies with social media filters. Bad back alley trash that could be turned for a profit by...
Andy Warhol famously created art pieces by remixing existing photographs and silkscreening them or otherwise remixing them. Does anybody seriously contend that Warhol's pieces aren't art? (They may, I genuinely don't know. I suspect not, though)
Raises hand.
Most of, and certainly almost all of his most famous pieces, are the art equivalent of selfies with social media filters. Bad back alley trash that could be turned for a profit by anyone with a folding table and a pair of stockings in NYC and probably were already doing so prior to Warhol. Andy Warhol is, like many allegedly transformative artists/people, a product of right time, right place marketing.
That's all only true if the only way you're using image AI is typing words in a box. It's true that's how many people use this technology now, but there are so many deeper ways of using the...
That's just something you can't do by typing a few words into a box. Creatively, it really is functionally equivalent to looking something up on Google Images. Maybe you can find something like what you had in mind, or maybe even something close to what you had in mind, maybe something that inspires you in a new direction, maybe you even find something you straight up like better than what you had in mind. But you can't use this type of AI to create something of your own, you can only sift and hope to stumble on something good enough.
That's all only true if the only way you're using image AI is typing words in a box. It's true that's how many people use this technology now, but there are so many deeper ways of using the technology that will become more standard soon. Tools for in-painting/out-painting allow images (AI-generated or not) to be edited, extended, and meshed together by AI with other images. Techniques like textual inversion and Dreambooth-fine-tuning allow people to give the image AI a couple example images of a person, kind of object, or even art style and have the image AI generate images (from prompts or for in-painting/out-painting) using what it's learned from the examples. Stable diffusion depth2img allows someone to generate an image that follows a depth map, creating an image taking place in a given 3d scene layout. There are tools that allow you to recreate an image with the camera placement changed: imagine designing an image of a city scene with a specific art style, and then "moving the camera" slightly to tweak the composition of the image before editing it more. Now imagine all of these tools in the hands of someone with skills and an artistic vision.
People are over-focusing right now on image AIs that just make an image from text with little human intervention. The best AI stuff soon is going to be the stuff that has a lot of human intervention. Even if it's the case that AI images with little human intervention are lacking in artistic value and human-driven meaning, that says little about the soon future of the technology.
A problem with this sort of speculation is that once you move beyond what the technology can actually do today, it gets very hard to make predictions. We aren't very good at predicting what clever...
A problem with this sort of speculation is that once you move beyond what the technology can actually do today, it gets very hard to make predictions. We aren't very good at predicting what clever researchers, implementers, and users will come up with. It turns into futurist punditry, which can be anything from "gee whiz it's going to be great" to "it's going to doom us all" and anything in between.
Investigating today's tech at least gives you something concrete to focus on.
Each of the things I mentioned already exist as open source projects and are often used by ML software developers working with image generation. All that's missing is for each of them to get...
Each of the things I mentioned already exist as open source projects and are often used by ML software developers working with image generation. All that's missing is for each of them to get packaged into applications that are usable to non-developers and work directly in more convenient image editing workflows. I don't consider myself to be speculating about whether these will be possible, though I guess I am speculating that these tools will create noticeably better results than plain text-to-image tools and become popular.
I think when you remove capitalism from the equation, a lot of the issues that arise with generative AI systems go away. I am not generally against automation but I do think it it's a repulsive...
I think when you remove capitalism from the equation, a lot of the issues that arise with generative AI systems go away. I am not generally against automation but I do think it it's a repulsive mess when it mixes with capitalism. On the flip side, it can possibly be used somehow as a tool for equity and opportunity too, so ultimately maybe it's a mixed bag.
For now though, in many cases when I encounter low-effort no-edit obvious AI art used in some form of project (like a music album) I just find it saps the quality and soul out of the project. Especially when it's used by entities that do have the money to afford better art, like big record labels. Capitalism will use AI as another way to cut company costs and cut corners across the board.
Maybe it's inevitable, but even as someone that has used and enjoyed AI art tools for personal use, I would never slap some random generated image that I "made" after 4-5 prompt refinements on an album layout and sell that as if I'm some talented designer. That just feels disgusting to me. I started out with a bunch of excitement about AI (because I'm a tech nerd so it excites that part of my brain), but I can't help but be soured on these things over time as I think more and more about them and react to how I see them used
I think there is value to the "heart and soul" or "effort" that goes into artwork (moreso for creative fields like art and less so for other areas of automation like robotics) and that generative AI essentially just, while helping some artists, will, just saturate those fields with tons of low quality "spam", basically.
and I generally don't really vibe with the arguments that the AI is just like humans in the way it "borrows" or "learns" from other artists' artwork. there's a difference between doing your own thing and being inspired by another artist, and me giving midjourney a prompt that says "make a painting of X in the style of <artist name>" and having it spit something out that looks a lot like that particular artist's work. The scale and the speed at which AI enables things like that, especially without consideration for the original artist, to me is a problem, not just something to hand-wave away as being okay because there exists a loose analogy for it. AI doing it and humans doing it are not equivalent. If it was restricted to only doing types of art or artists who are long gone figures of history, then maybe that's a different story. Lots of nuance here.
I get and have sometimes agreed with the comparisons to the complaints portrait artists had about photography, and so forth. I think there are some valid parallels to prior situations, but I still think these analogies are somewhat shaky. AI image generation is, IMO, an order of magnitude more powerful and an order of magnitude less human involvement than any previous paradigm shift in creative fields. Yes, for certain things a human has to guide the AI for a while to get a desired result. This is a bit of a lottery, though. I have gotten great images sometimes in the first 2-3 prompts, especially with v4 of midjourney. I have gotten images I probably could sell or use for designs and it not even be obvious that it's AI-generated. I think the effort that goes into photography, photoshop, and all other tools is just not on the same playing field as what AI enables. It's not just "another tool". I think it operates on a completely different level than that and I'm already exhausted and tired from all the BS that technology seems to bring us when it hits new levels of speed and scale (social media, etc) and I'm just not ready to embrace yet another avenue for tech to make us worse off
The more I hear about AI art, the more I feel like there’s a market for ethical AI. As in, artists can opt-in to have the bots trained on their art. Maybe even get paid for it.
The more I hear about AI art, the more I feel like there’s a market for ethical AI. As in, artists can opt-in to have the bots trained on their art. Maybe even get paid for it.
The cat is out of the bag. Even if the major models opt out of the training set, it’s possible for anyone to take the model and resume training it to incorporate the new data which takes not much...
The cat is out of the bag. Even if the major models opt out of the training set, it’s possible for anyone to take the model and resume training it to incorporate the new data which takes not much time at all compared to the base model.
There are also “style transfer” technologies where you can provide an image or some images and have ML extract the style and apply it over its new generated content. The research is already out there and people are doing this already. It’s too late.
I still think there’s a market for it, though not the largest market. Some people want to use more ethical software, myself included, and some even pay a premium for it. There’s plenty of ways to...
I still think there’s a market for it, though not the largest market. Some people want to use more ethical software, myself included, and some even pay a premium for it. There’s plenty of ways to steal music, and piracy of music used to be big, but once we made it easier to pay for and realized we wanted artists to get paid, a lot of folks started paying for streaming services. It was a disruption to the piracy scene.
Stable Diffusion users on Reddit have been discussing using SD without specific artist prompts, rejecting models overfit to artists' styles, and fine tuning on one's own art. I don't know if using...
Stable Diffusion users on Reddit have been discussing using SD without specific artist prompts, rejecting models overfit to artists' styles, and fine tuning on one's own art.
I don't know if using SD and fine tuning on your own art is ethical enough for you, as the base model was trained on such a large corpus, but it seems totally reasonable to me.
I won't be surprised if artists in the future sell models trained on their own art. Even fits a subscription model. But it probably will require a certain rapport with customers to keep them invested in supporting the artist in improving both their model and as an artist.
Related: i just heard a seasoned (but not at all technophobic) lawyer relate about having just negotiated a contract against an ai. He sid it was a tough negotiation. I’m trying to get more...
Related: i just heard a seasoned (but not at all technophobic) lawyer relate about having just negotiated a contract against an ai. He sid it was a tough negotiation. I’m trying to get more details but it’s fascinating.
Related: Justin Roiland's High On Life contains AI art and voice acting. Imgur: A bunch of the posters in the starting room of "High on Life" are clearly AI generated.
I've been using Stable Diffusion heavily for a couple months now, and I'm getting increasingly frustrated by the backlash and smear campaigns from people who are ill-informed about what this technology is and how it works.
This is pattern recognition software. When you train a model, the software ingests images and metadata about them. It looks for commonalities between annotations (image captions/descriptions) and between the images themselves. It assigns probability weights to those commonalities, meaning it comes to associate certain graphical characteristics with certain words or phrases (tokens). It's able to find more and richer patterns if the data set it's trained on is large and well curated.
This is a mathematical distillation of the same process human artists go through as they develop their skills and personal styles. Nobody picks up a paintbrush and creates a masterpiece on the first try. Great artists only become great by exposure to and study of the works of other masters that came before them. You don't grok the finer points of shading, perspective, composition, style, etc., without observing and comparing what others have done first.
This technology does not save images in a database and composite them together into collages. The Stable Diffusion model is trained on upwards of 100TB of images, but the model itself weighs in at a tidy 4GB. It's literally incapable of clearly reproducing any of the images it was trained on. What it can do is synthesize new images by starting with a field of random noise and iteratively modifying it until patterns emerge which correspond to the weights it was trained on, with its attention during this process probabilistically steered by user-supplied text prompts.
This is, again, a mathematical distillation of human inspiration. We learn about the world through observation, and the act of creating art is to transform a blank canvas into something unique that meaningfully represents the understanding we gained through those observations. If this is plagiarism, all artists plagiarize; everything is a remix. I can empathize with an artist wanting to withhold her works from a training set. But to me that's like a painter showing their gallery to another painter while saying "you may look at my paintings, but may not learn anything from them, lest they influence you into painting something of your own later on that I deem too similar to mine."
Now, after saying all that... this is a tool, and an imperfect one. Despite claims that "you just push a button and get a masterpiece," working with SD is an exercise in patience. It's laughably bad at understanding details like multiple subjects, relative positioning, scale, composition, and anything intricate including hands and feet, not to mention the barest understanding of basically any designed object. The flaws in the training data and annotations are readily apparent, it has blind spots a mile wide and produces freakish garbage more often than not. Any output with text in it is completely unusable. It defaults to 512x512 resolution gens and compute resources climb astronomically if you want more than that.
My workflow involves painstakingly crafting a prompt that's both detailed enough to produce something interesting, and general enough to sidestep the gremlins that lurk inside SD when you overspecify. I produce countless low-res gens as I go, checking and correcting my assumptions. Once I'm reasonably happy with a prompt I use it to generate 100-200 high-res images and then sort through those manually to find anything good enough to work with. Then I'll pore over whatever's left, manually cleaning up any remaining AI weirdness, janky hands, nonsensical backgrounds, oversmooth textures, bad cropping, and so on. I use a combination of AI inpainting, outpainting, and good old fashioned Photoshop. This is art. It's far from effortless.
And here's where I think artists need to get on board. These are incredibly powerful new tools for your toolbox. Just as the art world reacted to and then embraced digital bitmap and vector formats, and the innumerable analog innovations that came before them, so we need to recognize what this is and avoid throwing out the baby with the bathwater. In the hands of a skilled craftsperson, an AI toolset is a game-changer. I'd encourage all artists to play with the unified canvas in InvokeAI and see what they can do with it. It doesn't even need to involve new image synthesis, it's astonishingly good at manipulating existing images, for example restoring damaged photos or quickly blending in manual edits.
The fact is, Pandora's box is open now. You can't uninvent this technology and it will only get better over time. So do we dig our heels in and resist change? Or adapt and allow our own horizons to expand with it?
It took the subject of this article three days to make a children's book and start selling it on Amazon. Do you know the normal timeline for this sort of thing, making art, writing the text, coordinating between artists and publishers? A year, minimum. The people whose lives are going to be upended by this are angry because they're scared, and they're scared because this will completely change the industry and their lives. They don't care about any of the technical things you mentioned, just that the skill they have spent a lifetime developing (often through significant personal hardship, given how little most artists and writers made even before this) has been suddenly devalued. Why would anyone buying art care that there are flaws that need a little cleanup or that there's some work to choose between a hundred best attempts when Stable Diffusion and ChatGPT will take you 95% of the way there? Their lives will never be the same.
I will push back on the idea that artists will or should get on board with these trends. These programs are great for putting out something that matches imagination to existing artistic or narrative styles. But (at least for now) if you have a vision that doesn't match what's come before you've got to have a person involved. How will those people make a living creating art long enough to develop their own personal style? Where will new themes and visuals come from to feed the gristmill of automation? It's all well and good to automate art, but these programs may be the artistic equivalent of clearcutting a forest; if you harvest everything now, you'll stop new growth.
I need to recognize that "AI art will devalue existing artists and threaten their livelihoods" is a separate argument from "AI art is plagiarizing existing art." I was mainly intending to make a case against the latter. It seems that a lot of discussions about this online are conflating the two, which doesn't do anybody any favors.
I agree with the former point, actually. AI art is a threat to the art world as we know it today. The footnotes of history are filled with epitaphs of industries killed by innovation. The automobile decimated the livelihoods of farriers. The electronic computer made human computers obsolete. How many telegraphists were put out of work by telephones and fax machines? I don't intend to diminish any of those professions. These were hard-working, deserved people with specialized knowledge, who had the rug pulled out from under them by circumstances beyond their control. No one's hardship should be celebrated.
Still, somehow, we soldier on. Those who can read the writing on the wall are wise to adapt. They can carve out new niches for themselves. In many cases, instead of being crushed by a new technology, there are those who find a way to springboard off it to greater success. On the other hand, those who indignantly try to keep on as they always have will go down with the sinking ship. Again, I don't mean this with any ill will. I only mean that, in terms of career survival strategies, life sometimes imposes big choices on us, whether we want to face them or not.
As a side note, I've really enjoyed Jason Feifer's Build For Tomorrow podcast. Each episode pulls an example from history of a disruptive change, like the invention of the bicycle, and examines the naysayer response at the time. Spoiler: The same arguments are made every time. There's nothing new under the sun. This is at least an easier pill to swallow when viewed in the context of history instead of the tenor of in medias res flamewars online.
Self publishing, as the subject did, takes much less time. A friend of mine is an author for a living, he turns out 3-4 novels a year with cover art, solo writing and editing with an artist for the covers. I doubt the publisher is using up 9 months to get their end done, so if someone can't manage a children's book once a quarter then they're really freaking slow.
Should we lament the passing of town criers, scribes, carriage makers, blacksmiths, or the most obvious in regards to art and technological advancement: portraitists - almost entirely replaced by photographers (who have their own arguments to make on whether or not they are artists).
At one time art required physical media, paint, ink, chalk, and the like along with the training and skills needed to wield those tools. The rise of digital art, illustrator programs, and photoshop hasn't stymied art, it's given art the accessibility that nothing before it ever could have. How is AI generated art any different? It's a tool used to provide an output of the input given. A paintbrush outputs a slice of color based on the inputs of the hand that holds it.
And as with all things, there is ethical consumption. I choose not to eat factory farmed meat, caged eggs, or buy slave labor electronics. People that value art will do the same and choose not to purchase AI art and support those that share those values.
Since all art is derivative, humans artists are nothing more than biological AI, just as incapable as the AI of creating something truly new. We've just given names to the derivative styles that can all be eventually traced back to our origins. If prehistoric humans didn't slap their hand on a cave and spit dye at it we wouldn't have Monet. AI doesn't change the fact that to produce something truly new requires a human. If anything AI is giving humans access to more artists than ever before since it was trained on more artists/art than a single human could ever experience in a lifetime. Putting in a prompt has the potential to expose the prompter to more art styles than ever possible prior to now, it's a tool. One that can be used to inspire or create.
As for how an artist should make a living until they develop their own style (of the very few that ever do), they do so the same way anyone else does, by doing lower paying jobs until they have the skills necessary to demand high paying jobs. Actors, comedians, singers, etc. all have to work up to the positions they want doing things that are, more often than not, unrelated to their chosen profession. Why should artists or writers be any different? But in the same breath, artists especially have the advantage of patrons. No one (hyperbole) is giving Suzy the waitress a patreon subscription until she gets her big break, but there are thousands upon thousands of artists with such subscribers, supporting their lives wholly or until they can survive on the income of their wares alone. I'm one of those people that is a patron of artists doing so because I believe in their work. A friend of a friend is a real life patron of the arts, in the centuries-past definition of the word, of an opera singer; paying to supplement her life while she trains, travels, and performs.
Is there enough to go around for every person that wants to be an artist to live off of their art alone? Debatable, but unlikely in this world of scarcity.
Funnily enough for your analogy, forests require clearing from time to time for new growth to occur. If trees do not fall or fire ravage the valley then sunlight does not reach past the canopy and nothing new can flourish. Deviantart alone has 61million active users, even if we assume only 10% are actually creating anything, that's still 6 million artists churning "tens of thousands of original pieces of art every day". How much of it is even worth the hard drive space it takes up? Just because something is made doesn't mean it's worth anything. Perhaps AI is the forest fire needed to clear the canopy and let the sunlight through.
...yet.
Furthermore, humans are affected by their sociopolitical contexts and anxieties: war, racism, sexism, poverty, ennui.
And human creativity also refracts through human biological impulses: maternity and paternity, sexuality!, jealousy, and whatnot.
Can you define those terms further and maybe provide some rules that would let us test if a system is imaginative or not?
If we were looking at a sentient and imaginative system, how would we know?
The art world will get upended. But this isn't the end of new styles. At some point people will develop new styles by slowly creating a dozen images in the style and then fine-tuning a model on that style.
Technology has always revolutionized art. Photography pushed artists from realism to Impressionism and Modern Art.
Modern phone based cameras are revolutionizing photography.
Along the way artists who were invested in old ways of doing things struggled, and artists who adapted and changed thrived.
Yet there is still a room even today for photo realistic drawings, wedding photographers, portrait photographers, Photoshop experts etc...
I know it wont be easy, but I have faith.
I've been thinking about the role of art in society quite a lot over the years. While I'm at least in theory agreeing with you, I do have some observations that I would like to share.
When it comes to art of any sort, the actual value is entirely intrinsic. Which is to say that it doesn't actually have any monetary value. There are many types of art in the world but most people don't have the ability to appreciate the majority of them. For instance, sewing is an art, but most people look at clothes or other sewn fabric goods as art pieces; they don't appreciate the worksmanship and creative choices made when deciding which stitches to use.
One of the things that has historically been used to give value to any given piece of art is the amount of labor it took to produce it. And for a great deal of graphic arts, that's one of the most common ways that art gets valued; an artist receives a request to make something and then they consider their materials and time and give the requestor a price based on those. We've seen countless ways that this kind of arrangement has been broken, though. The development of photography killed off the business of classical portraiture because now there was a device that can capture lifelike images in seconds instead of weeks. Lace has almost entirely disappeared from fashion because it has transformed from a difficult time-consuming artform into something a machine can complete in a few minutes.
But at the same time we know that value can be subjective. That's why (in theory, if it's not just a money laundering scheme) there are pieces of art that be worth obscene amounts of money. And I think that understanding that is how artists are going to be able to drag themselves out of this situation. There is one thing that human artists have that AI does not, and that's intentionality.
I'm sure we all know that these types of AI are imperfect and they will probably get better over time, but they lack fundamental things that humans do. There's no dataset that will cover the entirety of a human's experience. They're also really bad at understanding what it is that you actually want; if you go even the slightest bit into the realm of imagination it will come up with strange deformed outputs. They're also really poor with accuracy; if you tell it that you want something in a specific space it won't understand you. Just for fun try asking it something you would expect a computer to be really good at drawing; try "a single black pixel" or give it an equation for a line. In a matter of speaking, the algorithms take inspiration from your prompt, but then they draw whatever it is that they feel like.
I expect we will always appreciate human performances, but in different ways than we do now. As technology advances, we have more choices about when and how we appreciate performances. We have more options and it becomes increasingly optional.
Historically, this happened via the invention of printing, the ability to record and replicate music, and the rise of the film industry and TV. A superstar can become a billionaire but the average artist doesn't earn much. You no longer need as many musicians to let everyone listen to as much music as they like, while charging nearly nothing.
On the other hand, the ability for millions to enjoy the same movie means that more money can be spent on them, so there are more specialists. All the people you see in movie credits get a job of some sort. (Or if you want lots of dancers, Bollywood has you covered.)
It can go the other way too. Big bands are expensive so you tend to see smaller bands these days, or just a DJ.
So there's a question about which human performances are the ones people care about and will pay for, and which ones are superfluous? Also, are you getting what you paid for? What counts as "cheating" depends on what people agree to as the rules of the game.
I wouldn't expect all artists to shun these tools. Some will surely defect, invent new art forms with new rules, and find an audience that appreciates them.
I hate that internet discourse descends into death threats and abuse so easily. It's not even productive or effective compared to just talking things out reasonably, like what?
Anyway, my first thought reading the discussion here is the contrast between AI art opinions here on tildes (demographics lean more tech) and on tumblr (demographics lean more creative) is so stark.
To be honest, I can agree with arguments about how the march of technological progress may come at the cost of livelihoods and of course, how capitalism negatively plays into this. But I don't really see any comments (yet?) on how this progress may come at the cost to culture if AI developers and regulators fail to protect artists' roles and relevance.
Artists' main concern may be their livelihoods but they also suffer through the carpal tunnel, demanding deadlines and clients, lack of opportunities and employment benefits like insurance... because they still love what they do. They enjoy both the process of creating and because their labour gives rise to beautiful, unique and impactful products that they can feel satisfied with, especially with the potential social audience and financial rewards. Keeping in mind even pre-AI, many budding and professional artists wouldn't reach half the audience or compensation their work deserves.
As a writer, any story I write could be reductively described as a "remix" of stories, tropes and culture that have existed for centuries. So I have to believe that my take on it, my version, my individuality, my perspective and my vision, will be just unique and different enough to be meaningful and valuable to some one else out there. At the heart of art is the human drive to make something they thought of and show it to some one who'll care about it. (Which will probably and hopefully remain possible even in a world with AI dominating the markets.)
But regardless, I feel like flooding the world with AI art (and writing) devalues that creative passion significantly in an already difficult and oversaturated market. Competing with AI turns the creative journey into an inefficiency, and art into a commodity. The role of the artist is reduced to that of a hobbyist or at best, a small-scale artisan, with limited visibility beyond that. The individual cultural origins and impact of an artist's work no longer matters to an average buyer or studio. Most artists not good enough to survive, would have to quit. Get a different job, maybe one in tech.
Because no matter how hard they work, an AI can do something similar, faster and cheaper. And when it's similar enough to be "good enough", the number of people who care that it's not traditional human-crafted art will be fewer. Traditional art itself will be harder to find, harder to see... harder to learn and harder to teach.
So, yea, I'm on the side of those scared to lose the jobs they love, because I don't want a world where we have fewer artists. Where the bar to succeed is set impossibly high and only certain artists with enough mass appeal will be able to keep creating for a living. Less originality and less diversity and less authenticity will only lead to a more and more uniform cultural landscape.
It makes me think of how we've seen CGI effects and animation start to dominate Hollywood indirectly because 2D animators and practical SFX designers are unionised. It makes me think of Disney's oligarchy on blockbuster movies that never challenge the status quo too much and how nothing seems fresh or original in cinemas anymore and yet mediocre movies continue to make billions. Technologically, they're excelling but culturally, how are they doing?
But maybe I'm wrong, maybe regulations will protect artists and AI will be an amazing tool of creation to strengthen artists' rights and working conditions, and not a weapon for more corporate exploitation. Maybe everything will turn out mostly okay and artists will be just fine. I'd love to be wrong!
Ahh, I'm glad! That's the main reason I find myself posting on Tildes, tbh, because I know there's a limited demographic active here but I like the culture and while I don't usually have anything to say, I hope my offering other perspectives is welcome.
And in fairness, I myself lack the full expertise to know how AI could be used in different ways, and reading this thread did enlighten me to the potential it has to elevate art in the hands of creatives and I truly think if not for the overwhelming context of capitalism (specifically in tech and in dealing with art & copyright & digital rights), there could be so much more good to come out of this.
Human generated art is too. No one produces an original work from nothing. The difference here is the ease of access. AI tools have opened art production up to everyone, not just those who spent years training (learning to derive existing work). It's understandable people with huge existing investment are threatened by the devaluing of that work.
Studios are going to lose their monopoly on production, and the influence the guilds have on blacklisting people who don't go along with their collective. That will probably generalize to all forms of entertainment, news, and propaganda too. Check out the AIs that generate 3D models and it's pretty clear that a photograph is all one will ever need to create an entire movie set or video game level in just a couple of years, and it'll take mere seconds to do either one.
Check out the list of AIs in the top comment on that video that can provide everything one needs to produce a video game. Now mix in the insane capabilities of ChatGPT and it won't be long before the AIs can do the coding too. That bot is already helping people with their PhD thesis and getting it right.
It seems democratizing, at least to me. I can sit here and create anything I desire without interference from another human being becoming necessary. I don't have to give up a single one of my rights or freedoms or sell out any of my principles to get investor money, or change things to placate some group of gatekeepers. I can sit down and tinker, and in a couple of months I've got an Avatar movie series or a Warcraft MMO or a new DC comic book series.
At that point, just sorting through and curating a million movies, a hundred thousand games, a billion paintings, and tens of millions of books a month becomes our new problem. What do you do when more masterpieces come out in a single month than a human can experience in their entire lifetime? I guess you get an AI to engineer us a longer lifetime. :P
The political implications of this are mindboggling, as it will become increasingly more difficult for any group, institution, or government to exert overt or covert influence on creatives as time passes. No gatekeeper can control that many gates opening that fast all over the world. Our grandchildren will probably have the same creative options as any vertical monopolist entertainment company has today, all from the comfort of their favorite chair in the basement - as long as they have the leather patience to guide the machines.
At that point, I guess it comes down to a simple question - do you have anything worth saying?
Tech is starting to speed up again - but this time, it's not just the hardware. AI is moving faster than ever and we're starting to see the various generations of any given narrow AI one-up themselves on the order of months, not years. Oh, and the need for large training sets is not necessarily going to remain a requirement for long. Our silicon is going to hold us back the most, but eventually that will give way to graphene, silicene, and photonics.
That new hardware will probably be designed by an AI as well, and it won't take the 60 years it took humans to iterate from the IBM 1401 to Threadrippers. It'll do it in twenty, perhaps less, and it's likely that newer hardware will also make inroads on quantum computing. Photonics is a natural fit for that task, as it satisfies more of DiVincenzo's criteria.
Kurzweil's prediction from the early 1990s of 2029 as a ballpark date for an AGI passing the Turing test and 2045 for a singularity-type event is starting to look dead on target now, not ridiculous. Capitalism has no brakes, either. The next two decades will likely be the most interesting and terrifying decades this planet has ever seen.
I'm happy for tech to be democratising! And I really appreciate your comment explaining how it could be so, I'm not familiar with the overall AI vision and the detailed projections and possibilities.
But I also see the reality of what tech has and hasn't done for democracy where we can barely control misinformation and hate (and that's if the guys in charge actively want to commit to it, here's looking at Twitter and Facebook) and regulations are too slow to enforce accountability so tech responds to markets primarily.
Here's an artist discovering a custom AI model producing art very similar to their art style, as a visual of who's on the other side of the road when the brakes on change fail. I think if nothing else, it's okay to grieve and empathise with the collateral damage being done.
I don't have anything else to add myself, I think your points are just as valid, it's a complex topic with a lot of grey nuance and ultimately, we'll each speculate on the future with as much optimism or pessimism as we individually feel inclined.
I find it hard to believe that someone can work in such a relatively small industry and not be aware of how much crap there is in it.
While there are a huge amount of amazing kids books out there, there are SO MANY terrible ones. At the worst end there are entire publishing houses which seem to exist purely to turn out cheap, rubbish cookie cutter crap. Those jobs are 100% going to go to people prompting ML systems, but they were worthless jobs anyway and it will probably make the books slightly better. But ChatGPT etc is not going to replace Oliver Jeffers or Lauren Child, Jon Klassen or Shaun Tan. ML still can't have ideas and this is the thing I think so many detractors don't seem to realise.
I don't entirely see how long something takes to make, or how much effort it involves should make any difference. Terry Pratchett used to turn out a couple of full length novels a year, didn't make them any less good. Meanwhile Harper Lee wrote two books over 55 years, one of which was a bit crap.
If it really does take three days to make a passable children's book that makes money, then it seems likely that there would be competition to churn out more books. (Much like happened with app stores.)
More likely, they won't make money, other than a few hits that go viral. And this is pretty much the case already for self-publishing.
A fifteenth century technopanic about the horrors of the Printing Press.
And people said the same thing about computers assisting in design and illustration. Back in his day, my ~65 year old graphic design professor had to paint letters on a can by hand for one of his projects. But there is a difference. When computers started getting involved, he was excited, just imagine how much more a designer or artist can do when they're not wading through the tedium and difficulty of executing tiny, fiddly details. But they're still being wholly created by a person who has to understand all of the same art fundamentals as every person before them regardless of the medium or tools they had available. Whether you're putting chisel to stone, brush to canvas, or stylus to digital tablet, there's a vision you're bringing into existence that requires hard-earned skill.
That's just something you can't do by typing a few words into a box. Creatively, it really is functionally equivalent to looking something up on Google Images. Maybe you can find something like what you had in mind, or maybe even something close to what you had in mind, maybe something that inspires you in a new direction, maybe you even find something you straight up like better than what you had in mind. But you can't use this type of AI to create something of your own, you can only sift and hope to stumble on something good enough.
They have their uses, for example one of my graphic design projects was pushed in a new direction by something I saw in a smeary AI soup. Another project was greatly helped when I was struggling how to even represent dirt in an Art Noueveau-like style. I couldn't find any references and everything I was doing looked awful, but a few rounds of harassing an AI gave me something I thought I could work with. But at least right now they're not the next technological advancement in creative tools. Because you really can't use these to create something that's yours. It's just an endless Google Images search on shrooms.
I have a lot of thoughts about some of this stuff. Forgive me if I ramble...
There are lots of kinds of art. Some folks really like Jackson Pollock or other types of non-representational modern art. In fact my sister-in-law is very into acrylic pours and other types of abstract art. I would argue that art pieces like Pollack's, or acrylic pours, are as much created by basic physical processes and random chance as they are by a person's skill. You have to know how to set up the process, and there's a certain amount of experience that you can leverage to make sure the outcome is more pleasing to the eye, but at some point gravity and fluid dynamics are doing as much or more of the creating as the artist. Yet I don't really see anybody arguing that those pieces aren't art just because most of the actual composition is done by gravity.
Andy Warhol famously created art pieces by remixing existing photographs and silkscreening them or otherwise remixing them. Does anybody seriously contend that Warhol's pieces aren't art? (They may, I genuinely don't know. I suspect not, though)
And none of those things invalidate other art forms that require more human input or human skill. Traditional oil painting still exists. Sculpture still exists (alongside 3d printing). Scrimshaw and weaving and hand sewn fabric arts and any number of primitive, skill and labor intensive arts and crafts still exist even though we have made it very simple to create most of those things with machinery and computers.
I do think there is danger than ML and automation are going to cost people jobs. In fact I think that automation is going to be a threat to a lot more than just professional artists' livelihoods in the coming decades. We're going to need fewer lawyers and paralegals to create boilerplate contracts. There will be less demand for copy writers, programmers, cashiers, call center workers, and God only knows who else. Sure, automation and machine learning will create a few new jobs, but not as many as it causes to be lost.
I think we're in for upheaval for a lot more than just artists and designers. If we react correctly to what's coming, if we create social safety nets and redistribute the wealth that is created by automation, then I think people will need to work less and will have the time and the peace of mind to create things like art, music, and writing just for their own sake and not so that they can pay rent and buy food. I'm not optimistic, though, that our political systems are capable of working toward that kind of collective long term good. I hope I'm wrong about that.
I don't think anyone doubts that this will cost jobs. However, it is possible that, in the future, AI art will be considered an overall improvement and a fundamental aspect of our lives.
I actually agree with you, at least on days when I'm feeling more optimistic. And full disclosure, I love AI image generators, I've been in betas for several of the models that are out right now and I spend a good bit of time playing with Stable Diffusion and various models that have been built on it. I think that ML and automation could bring a lot of positive things in the future, but I do worry that there are going to be some pretty bad growing pains. I'd love it if we could be forward looking enough as a society to see how mass automation could and probably will affect us and put safety nets in place for the folks who will be the most vulnerable, but I'm not always optimistic that we will be.
Everything will eventually settle, but I agree with you that it is unfortunate that some will have to pay more than others. There should be mechanisms in place to accommodate those harmed by such changes. Universal basic income would help.
Commercially speaking, there was a recent copyright dispute between the photographer, and Andy Warhol's estate.
The Supreme Court foundA judge found that Andy Warhol's screen prints were transformative enough to not be copyright infringement. but was overruled.Edit: nothing has been found, I fail at basic reading comprehension.
Worth noting that the Supreme Court hasn't actually ruled on that case yet, no opinion has been issued.
when you are right, you are right.
Raises hand.
Most of, and certainly almost all of his most famous pieces, are the art equivalent of selfies with social media filters. Bad back alley trash that could be turned for a profit by anyone with a folding table and a pair of stockings in NYC and probably were already doing so prior to Warhol. Andy Warhol is, like many allegedly transformative artists/people, a product of right time, right place marketing.
That's all only true if the only way you're using image AI is typing words in a box. It's true that's how many people use this technology now, but there are so many deeper ways of using the technology that will become more standard soon. Tools for in-painting/out-painting allow images (AI-generated or not) to be edited, extended, and meshed together by AI with other images. Techniques like textual inversion and Dreambooth-fine-tuning allow people to give the image AI a couple example images of a person, kind of object, or even art style and have the image AI generate images (from prompts or for in-painting/out-painting) using what it's learned from the examples. Stable diffusion depth2img allows someone to generate an image that follows a depth map, creating an image taking place in a given 3d scene layout. There are tools that allow you to recreate an image with the camera placement changed: imagine designing an image of a city scene with a specific art style, and then "moving the camera" slightly to tweak the composition of the image before editing it more. Now imagine all of these tools in the hands of someone with skills and an artistic vision.
People are over-focusing right now on image AIs that just make an image from text with little human intervention. The best AI stuff soon is going to be the stuff that has a lot of human intervention. Even if it's the case that AI images with little human intervention are lacking in artistic value and human-driven meaning, that says little about the soon future of the technology.
A problem with this sort of speculation is that once you move beyond what the technology can actually do today, it gets very hard to make predictions. We aren't very good at predicting what clever researchers, implementers, and users will come up with. It turns into futurist punditry, which can be anything from "gee whiz it's going to be great" to "it's going to doom us all" and anything in between.
Investigating today's tech at least gives you something concrete to focus on.
Each of the things I mentioned already exist as open source projects and are often used by ML software developers working with image generation. All that's missing is for each of them to get packaged into applications that are usable to non-developers and work directly in more convenient image editing workflows. I don't consider myself to be speculating about whether these will be possible, though I guess I am speculating that these tools will create noticeably better results than plain text-to-image tools and become popular.
Yep, that's true. I should read more carefully.
I think when you remove capitalism from the equation, a lot of the issues that arise with generative AI systems go away. I am not generally against automation but I do think it it's a repulsive mess when it mixes with capitalism. On the flip side, it can possibly be used somehow as a tool for equity and opportunity too, so ultimately maybe it's a mixed bag.
For now though, in many cases when I encounter low-effort no-edit obvious AI art used in some form of project (like a music album) I just find it saps the quality and soul out of the project. Especially when it's used by entities that do have the money to afford better art, like big record labels. Capitalism will use AI as another way to cut company costs and cut corners across the board.
Maybe it's inevitable, but even as someone that has used and enjoyed AI art tools for personal use, I would never slap some random generated image that I "made" after 4-5 prompt refinements on an album layout and sell that as if I'm some talented designer. That just feels disgusting to me. I started out with a bunch of excitement about AI (because I'm a tech nerd so it excites that part of my brain), but I can't help but be soured on these things over time as I think more and more about them and react to how I see them used
I think there is value to the "heart and soul" or "effort" that goes into artwork (moreso for creative fields like art and less so for other areas of automation like robotics) and that generative AI essentially just, while helping some artists, will, just saturate those fields with tons of low quality "spam", basically.
and I generally don't really vibe with the arguments that the AI is just like humans in the way it "borrows" or "learns" from other artists' artwork. there's a difference between doing your own thing and being inspired by another artist, and me giving midjourney a prompt that says "make a painting of X in the style of <artist name>" and having it spit something out that looks a lot like that particular artist's work. The scale and the speed at which AI enables things like that, especially without consideration for the original artist, to me is a problem, not just something to hand-wave away as being okay because there exists a loose analogy for it. AI doing it and humans doing it are not equivalent. If it was restricted to only doing types of art or artists who are long gone figures of history, then maybe that's a different story. Lots of nuance here.
I get and have sometimes agreed with the comparisons to the complaints portrait artists had about photography, and so forth. I think there are some valid parallels to prior situations, but I still think these analogies are somewhat shaky. AI image generation is, IMO, an order of magnitude more powerful and an order of magnitude less human involvement than any previous paradigm shift in creative fields. Yes, for certain things a human has to guide the AI for a while to get a desired result. This is a bit of a lottery, though. I have gotten great images sometimes in the first 2-3 prompts, especially with v4 of midjourney. I have gotten images I probably could sell or use for designs and it not even be obvious that it's AI-generated. I think the effort that goes into photography, photoshop, and all other tools is just not on the same playing field as what AI enables. It's not just "another tool". I think it operates on a completely different level than that and I'm already exhausted and tired from all the BS that technology seems to bring us when it hits new levels of speed and scale (social media, etc) and I'm just not ready to embrace yet another avenue for tech to make us worse off
The more I hear about AI art, the more I feel like there’s a market for ethical AI. As in, artists can opt-in to have the bots trained on their art. Maybe even get paid for it.
The cat is out of the bag. Even if the major models opt out of the training set, it’s possible for anyone to take the model and resume training it to incorporate the new data which takes not much time at all compared to the base model.
There are also “style transfer” technologies where you can provide an image or some images and have ML extract the style and apply it over its new generated content. The research is already out there and people are doing this already. It’s too late.
I still think there’s a market for it, though not the largest market. Some people want to use more ethical software, myself included, and some even pay a premium for it. There’s plenty of ways to steal music, and piracy of music used to be big, but once we made it easier to pay for and realized we wanted artists to get paid, a lot of folks started paying for streaming services. It was a disruption to the piracy scene.
Stable Diffusion users on Reddit have been discussing using SD without specific artist prompts, rejecting models overfit to artists' styles, and fine tuning on one's own art.
I don't know if using SD and fine tuning on your own art is ethical enough for you, as the base model was trained on such a large corpus, but it seems totally reasonable to me.
I won't be surprised if artists in the future sell models trained on their own art. Even fits a subscription model. But it probably will require a certain rapport with customers to keep them invested in supporting the artist in improving both their model and as an artist.
Related: i just heard a seasoned (but not at all technophobic) lawyer relate about having just negotiated a contract against an ai. He sid it was a tough negotiation. I’m trying to get more details but it’s fascinating.
Please share if you can get more detail. I'd love to hear about it.
Add me to the list of very interested parties @NoblePath.
My guess is it'd look something like this.
Related: Justin Roiland's High On Life contains AI art and voice acting.
Imgur: A bunch of the posters in the starting room of "High on Life" are clearly AI generated.