I have a degree in Studio Arts in addition to being a software developer (and recently I've started making a lot of creative work with AI). Back when I was in school, there was already plenty of debate about 'what is "Art"'? It's always been hotly contested. Most of my teachers felt that 'technique' was not Art. Others were more purists – life painters, etc – and felt that, no, technique is art.
Duchamp poked the art establishment in the eye with his 'readymades' (urinals on the gallery wall, etc.) Warhol with his 'factory' making silkscreens.
The 'reverse Monet' (too lazy to link) that caused such an uproar when it was shared last week was ironic – apparently people aren't aware that the impressionists themselves were considered 'not artists' by a lot of people at the time.
The camera and photography made many painters sit up and say, 'hey, it can't just be about "image" anymore' – enter abstract expressionism, fauvism, cubism, etc. Work that shifted the focus from the image represented in the painting ('a window into a world') onto the surface of the canvas itself: Rothko, Albers, etc.
The thing is, people have wild misconceptions about what this technology does. Many clearly think it's a photocopier. As if StableDiffusion is hiding copies of all these works inside. When in fact what's been done is to take all of this online knowledge and create an incredible new tool with it. Like a music synthesizer times a million.
It's understandable why this technology makes people uncomfortable. There are legit fears about displacement, economic uncertainty, etc. But we are fundamentally tool makers – it's arguably intrinsically part of what makes us human (borrowing from Carl Sagan's "Dragons of Eden").
If I were back in the art department last century, I'm pretty sure we'd be really excitedly discussing how these new tools will enable entirely new forms of expression. And sure, there'd be the old guard saying it was awful, and 'not Art'.
This is a really interesting debate – I just wish it scratched deeper than the surface more often than it does, with a bit more historical context.
> Many clearly think it's a photocopier. As if StableDiffusion is hiding copies of all these works inside.
In my latest series of tweets [1] I tried to argue that this is not the case in a less technical way, not sure if I was successful though. For anyone not fluent in Korean, my argument was as follows:
Stable Diffusion was trained with a data set composed of about 2.32 billion text-image pairs. The model itself by comparison weighs about 4 GB. If the model somehow stored those images (completely disregarding texts) in the compressed form, it would be less than 2 bytes (EDITED) per each image. Whoever downloaded an image from the internet can see this is absurd. In reality 4 GB of data is the juice extracted from the data set, and if you could actually recover an original image from it then the juice must have had a recognizable residue, which would be pretty undesirable.
Not a good argument, because this type of AI is basically exactly a compression algorithm - it’s the most efficient way we know to compress 2 billion image/text pairs down such that it can recreate the average of all the inputs retaining as much layered knowledge as it can.
It’s a very very fancy compression algorithm, but it still is one.
There is an important distinction between lossy compression and lossless compression. My information-theoretic argument shows that models couldn't have been losslessly compressed, and the juice analogy implies (but does not show) that there is no obvious way to "decompress" the original data set. I intended that, if both arguments are true, those models can't be a photocopier because they don't contain copiable works at all.
The model size:training image ratio seems more like an implementation detail more than anything.
The fact that people are writing prompts with named artists, and phrases such as "Trending on Artstation"- yeah, I'm not sure if you can just handwave that away.
“Trending on Artstation” is a genre curated just like Bauhaus or Romanticism. Just the wider populace votes instead of old noble families and wealthy bankers.
I don’t see the moral significance of the former genres versus the latter
In some ways it’s just a coincidence that those prompts do something. You can use names of artists that don’t exist and those work just as well, once you’ve figured out what they mean.
Also, even if the model didn’t come with knowledge of an existing artist someone could fine-tune it in, and it’s possible the model can learn about your art style without seeing any of your images anyway.
Others have alluded to this, but I'll make a larger claim: human artists actually are doing very little that is original. The styles, composition, and subjects are all derivative works. That means their "essence" can be recreated without them.
Another fascinating consideration, if artists names were excised from future models, either by law or by choice, the "mirror" artists (fake) could be the ones which become famous and the humans forgotten.
AI could become more generalized or there may be millions of models, doing specific things, strung together like with APIs. Either way, if an artist or trademark owner chooses some kind of explicit blocking, from the input side, it is possible that it would be the equivalent of your keyword censored from Google's index, Google Maps, Amazon, and so on. Disney? I don't recognize that word.
It might not be storing all of the images, but it's clearly storing copies of some of the source images.
A few weeks before OpenAI made DALL-E 2 open, I went through a series of prompts using well known artists as a basis. Vermeer was one of them. It generated some pretty amazing works that were 'inspired' but not direct copies of Vermeer's works.
I started feeding the same prompts into StableDiffiusion (the HuggingFace online demo) and it output direct copies of Vermeer's works with minimal modifications.
Do you have the exact prompt that essentially generated Vemeer's works? Omitted in the parent, but I also noted that while it is hard to reconcile the idea of human-directed transformative use with ML models, it might be actually easier to verify the models' plagarism---which is not equivalent to copyright infringement but can be a problem by its own---once you have a right prompt as they are decidable.
It's reproducing the folds in the headband, riffles in the clothing, etc. It's not just a "similar" reproduced image, it's clearly some minor variation on the original.
But I think this sort of thing is the exception rather than the rule (similar to Copilot regurgitating the text of software licenses). Girl with a Pearl Earring is so famous that there are countless copies of it on the internet. I don't know what kind of deduplication process was used for Stable Diffusion, but if you consider multiple photos of the same painting, plus images distorted by compression, or cropped, or embedded in other images, plus attempts to redraw it or even parody it, I can easily imagine that the training set contained thousands of images or more that are essentially the same work. And so the model memorized it, albeit at low resolution.
Similarly, it's seen countless different images of Pikachu.
However, this seems to only apply to the very peak of popularity. I tested some other Pokemon: it does a decent job with Charizard, but with Bulbasaur and Squirtle and Jigglypuff its output only somewhat resembles the character design, and for less popular Pokemon there's no resemblance at all.
> the exception rather than the rule [...] . Girl with a Pearl Earring is so famous that there are countless copies of it on the internet
I think that's part of the point, though.
Even if it's working like some kind of feature detection - "this is what 'nose' looks like [nose1,nose2,...,nose9999]" and "this is what 'girl with a pearl earring vermeer' looks like [vermeer1,vermeer2,...,]"
Clearly it is storing copies of those images internally, and constant re-enforcement of it finding noses and vermeer paintings in the sources pushes the priority of those features up so that it stores them.
We can't trust these models, as they exist right now, to reliably generate new works, rather than spit out something that already exists.
Technically speaking, there can be only so many such reproducible works in the model due to the information-theoretic constraint. But yes, I agree that this is a close enough reproduction of the Vermeer's work, and while this is indeed in public domain, you are right that other copyrighted works may be present in the model, in verbatim and extractable with a right but currently unknown prompt.
The photos of the painting on the internet aren’t original works either. This system is learning from everyone taking pictures. If you paid a very good artist to paint that painting, they’d work from a picture. No artist emerges fully formed without impact of what they’ve seen.
There’s a difference between an artist and a machine (for now), but it’s not unreasonable to assume that some internalisation of others’ work is fair.
Not reliably, no. But at least for this size of model, its capacity to memorize seems to be limited to works that are so popular that 'everyone' knows what they look like. And it also seems to generate them only when asked. So in any situation where a human is reviewing the output, it doesn't seem like there's a big risk of plagiarizing something by accident. Larger models may differ.
I feel like this is one of those occasions where perfect is an enemy of the good. Yes there might be some, very very famous, artworks stored inside it. But unless you think StableAI has completely blown past the current hutter prize winner, there is no way this thing is storing many things. The things it might be storing are too public to really be that big of a deal. It will spit out a Vermeer, but your submission to Divient Art in 2007 is definitely protected.
> there is no way this thing is storing many things.
I didn't claim that it was storing all of the images.
It is, however, quite clearly storing at least some of the original training images. What that definition is, I don't know.
To go back to the original point of this thread:
> The thing is, people have wild misconceptions about what this technology does. Many clearly think it's a photocopier. As if StableDiffusion is hiding copies of all these works inside.
I don't think it's a wild misconception, or even an unreasonable concern to be worried that it could output some of it's training images. As an end-user of the technology, you have no idea.
There's different sorts of art for different purposes (commercial, fine, etc.), and for many of these sorts of art having an AI make it would be acceptable.
But when I think "art" without a qualifier, I think "fine art". And the purpose of fine art, for me, is human communication of ephemeral things that aren't possible to communicate in language.
I don't want AI generated fine art because it turns the art into just a pretty picture. It's no longer about another person communicating something significant and human to me. It removes its purpose for existing.
In that view, I totally understand why some art communities would ban AI-generated art. I don't see a problem with it. I'm sure other communities will not ban such art. That's fine, too.
But I know which community I would be more interested in paying attention to.
> It's no longer about another person communicating something significant and human to me.
There is a person thinking about the prompt, typing in the prompt, painstakingly selecting the image that matches what they imagined in their head.
It just less elitist, as now even a child can create a complex, meaningful and beautiful images by leveraging the essence of human art (compressed into a single ML model).
I, for one, never had the time & chance to learn painting. Even if I did it as a hobby, it would take me up to a decade to become proficient at it.
And now I can ask AI to paint what I dreamt about last night (!!!), or create an impressionist painting almost exactly like I see it (or even better).
Fun factor is also there: I can see how an anime character could look in real life, or put a movie character in a new context , or just generate cute corgis wearing funny hats and driving cars
> There is a person thinking about the prompt, typing in the prompt, painstakingly selecting the image that matches what they imagined in their head.
Which does make this a type of art. It just isn't the same form of art as digital painting, the same way painting a tree is not the same type of art as growing and 100 trees, choosing one of them and taking a photograph of it. Photography is definitely an art-form, but it's a different art-form from digital painting, even if the end result of the both of them can be a digital image. Not better or worse, but definitely different.
> There is a person thinking about the prompt, typing in the prompt, painstakingly selecting the image that matches what they imagined in their head.
True, which makes it more akin to photography. I'm not saying it's "not art" (I'm saying the exact opposite of that), but that it's not art that is of interest to me. Much like my interest in art photography is very limited. Different people have different tastes.
> It just less elitist
I disagree. I think this whole subject is orthogonal to the concept of elitism. Sure, there are certainly art snobs who take an elitist perspective, but they're a very small segment of the art world.
Can't help but think about the obvious: maybe we should just skip all the drudgery of the natural convulsive progress and ruthless competition, push "accelerate" and get to the mature AGI civilization regime as soon as possible?
And there we should just ask our new creative oracle machines the obvious: "devise an understandable minimally invasive method of uplifting our own cognition and extending our biological lifespan". Surely at least some gene/cell/whatever therapy which does what we need is theoretically possible - the human beings aren't special enough for this not to be the case.
At least that's the zeitgeist I perceive among others who understand the significance of the unfolding progress.
Pushing "accelerate" costs a great deal. Investors will want a return on their capital: either in money pressed from unequal access to the resulting paradise or in direct power over this new god.
With billions upon billions spent on Big Science, asking the government to spend just 1 billion on an AI training run with guaranteed spectacular results doesn't sound too outlandish.
Why should we pay taxes, if the government won't even train AI for us, given such opportunity?
We all know that analogies from Hollywood classic scifi movies aren't productive or even at all useful in our mundane reality. The general-purpose AI is going to be a very useful, obedient, capable tool, and the great powers are going to seek ownership over it, like they do with various military and space technologies.
AI safety discourse is mostly still in 2000s, with outdated intuitions informing loads of rigid concepts specific to this time and community.
Large language models trained on cross-entropy loss for next token prediction aren't going to be dangerous per se, even in the limit - simply due to the nature of the objective and the distribution they are set to approximate.
We should make sure these powerful tools don't fall into a small exclusive set of power-hungry interests, though. Transformative tools such as these should benefit all of humanity and let it prosper.
If large language models trained on cross entropy loss for the next token prediction in the limit aren’t dangerous, then they aren’t enough by themselves for AGI.
An AGI, by virtue of being at least as general as a human, would be able to effectively form a plan for how to act in the world to achieve fairly general goals, in fairly general conditions.
Even if it isn’t agentic, if a program can formulate these plans in a way people can use, then the combined program+user still implements the optimization process.
That's why we have to make sure the user (committee) is safe and aligned. The usual silicon valley tech execs, or east coast financial bros aren't going to cut it this time.
The democratic solution here looks better than the expected rule of the (very) few.
It seems to me that the human effort spent in creating the original images in an AI's data set is an important element to consider when thinking of the "difficulty" here.
Certainly, but those human artists were also trained on a data set their whole lives. Raise a baby with no training data set and they might produce a handprint on a wall or a rectangle of facepaint. There's a reason indigenous tribesmen don't spontaneously invent hyperrealism.
However, there's a major misconception about training of humans and AI.
Image generating AIs are trained with massive amounts of images and text, however image generating humans train with much broader spectrum of experience.
Also, feeding a model tons of images created by humans (directly or by proxy) and claiming that AI is generating something completely new is a bit naive IMHO. Humans mix a much more broader and deeper experience pool to create things without prompts.
An AI model blurts out something derived from a corpus of images and text created by humans, that's all.
The technology is impressive for sure, and it marks a new era in terms of possibilities, but it doesn't take my breath away, sorry.
Further to your point, comparing humans to AI is not just misunderstanding AI, but also looks past how humans create and choose the styles which this AI is now reproducing through a statical approach. Without the human guidance the AI would be bland, AI is limited to the walls of whatever it has been exposed to - humans are not.
The AI emulates. Humans create. - The significance here is not trivial.
Humans emulate as well, and produce bland art as well.
I've seen lots of human-made art (especially on Artstation) that failed to elicit any emotional response in me.
The reason people are having this discussion is a fear of AI creating aesthetically pleasing and deep emotionally through statistical approach.
And the deeper fear that AI can potentially "hack" our senses by providing us with exactly what we want to see, read or hear (statistically).
i.e. if some subset of art is pushing our biological buttons more than all other art, and the effect is generalizable over human population, what would that say about us?
Well put. AI is an evolutionary death, in a sense. True Intelligence is not, at the point at which it is not, it is no longer artificial, and then ethical concerns must apply (rights for so-called AI citizens, etc.)
Humans mix a much more broader and deeper experience pool to create things without prompts.
That wide variety of experience changes the decision of what to create, but I think it has less influence on the actual output than one might expect. A person can have enormous life experience but if they have never seen a logo, a photo, a painting, a drawing, or any other kind of rendered image they will be no more sophisticated in their artistic output than a caveman. Modern artists stand atop a mountain that their ancestors had to climb inch by inch.
Actually, I don't agree. I take photographs and design logos for my pet projects sometimes, and I optimize for feeling most of the time.
I want my photos to make the viewer feel a particular way when viewing what I took. This affects colors, style, and everything.
Arguably, as long as the medium carries the message the creator intended, it's equal in my view. It can be a photo, a drawing, a digital painting or a physical painting with oils. They can create the same feeling if the artist aims for it.
This "optimization for feelings" is a result of my own non-photographic experiences. Sometimes a song, sometimes an event I went through, sometimes an art piece I saw on a medium I don't work.
From my understanding and observation, I saw that many artists work that way. They reflect their emotions in one domain (generally personal life) to another domain (the art they create). Also, there was a video which I fail to find over and over which shows how three designers got affected by the things they saw (and deliberately planted into their minds slowly) during a 20 minute car trip.
Yes, we're building upon this great mountain of experience and knowledge, yet our output is affected by what we experience in other parts of life, unrelated to the art we create.
Consider the following experiment: The music you listen changes your mood, your working speed, and what you feel; effectively changing how you operate and what you create. Exclude work from this, which is by definition something you have to do, unless you're literally dying.
"yet our output is affected by what we experience in other parts of life, unrelated to the art we create."
Right, and what happens when you let that loop feedback to itself?
An asphalt road on every corner, so we build more asphalt roads, we design only what has been designed before. Amplify this thousands of time through computer generated design, the feedback loop is closed and unchanging. Unless your goal is actually to create an independent organism (different moral issue), you are creating sophisiticated feedback loops, not worthwhile content generation. Unless by content generation you mean endless remixing...which is not the same.
Sort of like changing the color of a 3d model and claiming it is a new race with new attributes, as is often done on the cheap e.g. in videogames.
> claiming that AI is generating something completely new is a bit naive IMHO.
I don't see how AI art is less new than the vast majority of human art. Both can create unique compositions that are still deeply rooted in patterns and principles thousands of years old.
> Humans mix a much more broader and deeper experience pool to create things without prompts.
AI does not have understanding, but it's stealing the underlying patterns that are the end result of that human experience.
Absolutely. Though I think that speaks to one of the ways we tend to discount the difficulty of developing human artistry, rather than a defense of AI gen being "not hard".
Yes, that 4 GB of data can't be really independently generated, but the fact that a large chunk of that effort can be compressed into 4 GB is still amazing.
It doesn't compress each image separately. Rather, it compresses its entire training set into one dense region of cartesian space. Each pixel of each training image is in there, scurnched up next to every other pixel of every other training image. It's like sardines in a can, except in thousands (?) of dimensions. You have no hope of calculating how much space it needs for that just by dividing the size of the model by the dimensions of images in pixels, like you did.
You can see that it has memorised all its training material very clearly in its ability to perform style transfer. You can ask it to geneate an image of Pikachu in the style of Vermeer. It can do that because it has memorised images of Pikachu and images by Vermeer in its weird compressed form and it can combine them by finding a point on a gradient between their representation in its memory.
If it didn't have memorised images of Pikachu, or images by Vermeer, how do you guess it could generate images of Pikachu or in the style of Vermeer? It's not magick.
Better analogy: suppose you take 10 pieces of A4 paper and scrunch them up into 10 separate balls. Now suppose you scrunch them all up together into one ball. One manifold, taking ten times less space.
In the same way, you don't need to store each pixel of each image separately, or in its own sub-space. Yes? It's really good compression, that's all.
I've used the words "stock photography" in an image and saw the shutterstock logo across the image in the output. Not sure how that happens but try it out.
As a non-artist whose compensation is, so far, unaffected by ML-generated images,
I'm more interested in the effects the technology has on people.
Over the past 150 years it became possible to compute cube roots in milliseconds, instead of minutes. Slide rules became obsolete, and the things we learn to do do arithmetic changed.
Over the past 20 years, the way we look up arbitrary facts changed. We learned to construct Web search engine queries and we learned to evaluate the quality of results. Other library skills are still relevant, but certain kinds of rote memorization become obsolete.
Now people are rapidly learning a new "prompt language" which uses English words but combines them in a way which is for now something like computer programming (given an input string, compile it to an output image). Will "prompt languages" for text / image / video generation systems turn into a full-fledged, non-English language that you can learn?
What will be the social effects of people learning and using this language to communicate?
I think it will be interesting, but not result in total annihilation of the existing field. As a general counterexample, TV did not kill radio despite it having more dimensions for creativity. Radio changed, but it did not die, and was quite comfortable for many decades after.
Coallescing on prompt languages is probably unlikely, if only because part of the thing that defines a language is that a community has rules, both opaque and explicit, about what is acceptable in the language and what is not, and there's no effort to standardize inputs.
sign makers were put out of business en masse at a certain point, and there were quite a few, and it is high-skill, high-commitment work. Some kinds of print arts related to magazines for example, were next.
There is a chapter in the Stanford AI Index (2021) that opens with "The rise of AI will inevitable raise questions of how much these technologies will impact business, labor and the economy in general .. "
there are winners and losers here, not just interesting questions. In the 1930s America there was mass unemployment among many skilled "workers" .. can't imagine the 1990s ex-Soviet economies.. I dont think these questions are neutral or innocent.
And vintage sign making is now highly valued (whether hand-painted or digitally produced simulcra)
We tend to value scarcity. Things that resemble AI art will become passé and anything that is difficult/impossible for AI to produce will become highly regarded. I'm very curious to see how this shift in popular and/or high-brow taste will play out.
thousands of small companies sustain and employ humans, reliably now as sign makers? I live in a metropolitan area, sign makers were reliably found and nearby in my living memory. There are very few now, mainly as corporate franchises with catalog contents. How is the value of scarcity you refer to, being circulated in terms of reliable income and stability for humans living now? How plural are the products in modern markets
This assumes adjustment is even possible. Prior history suggests "adjust" is a euphemism for pivoting to unfulfilling lower wage work. If that seems like hyperbole to you I highly recommend taking some time out to research the before/after effects of globalization on rural factory towns in North America or the impact of high intensity corporate agriculture on family farms. If you want the tl;dr, just look up a graph of farmer suicides over time.
I mean, that has mostly been supplanted by graphic design as a field, which is still making signs, just not necessarily literally handmaking signs out of outdated technologies like neon tubes. And now also helping design websites and other such new technologies.
> In the 1930s America there was mass unemployment
Sure, but these are temporary disruptions. Now it's true that the pace of change feels immensely faster than ever before, and I believe that's an observable fact. But we adapt. The incentives to do so are, sometimes, life or death. When faced with that kind of situation, some small % will sink into depression, perhaps never to climb out. I'm sure that some online magazine (which surely used to be known for its print version, but is no longer...) will write an extremely long piece full of anecdotes about the plight of the artist whose livelihood was destroyed by the evil AI.
But I assert that it will always be a tiny percentage of the whole, because the vast majority of people like to eat food and live in houses. Many of them will look at these new tools and make amazing new things. Many will go do other things. That's life, man.
I had to think twice about this reply .. I believe that from a psychological, character development point of view, what you say is reasonable.. adjust to change and apply new skills in new ways.
However, there are different "lenses" through which one might examine large topics, and one lense might be that of personal challenge, adjustment and endeavor; but another lense is closer to The Economist Magazine, where factual snippets of market behavior, participants and results are traded every day, every week, every year. Any college educated person ought to be able to say, there have been real, serious and long-standing economic changes where thousands and millions of capable, good-enough people, had serious, years-long hard times up to and including starvation, war, and abundant death. Those without personal experience of that, or a close relative or similar imprinting, may not really consider this real. I had to learn it from books myself based on where I grew up. Others reading these words, know it very well.
Hand-craft preservation is a thing, I have heard.. so there is certainly a broad spectrum between "no more blacksmiths downtown" to "I send my print jobs via phone for pickup near the metro at an automated kiosk". It is said that nobody has a right to a job. However, The Economist Magazine exists for a reason, and things are not normal where I live.. Welcome to the New Not-Normal, as Jerry Brown said..
Arguably, that's basically directly saying that nobody has a right to life.
You either ensure the basic resources necessary to live are free, or you want everyone to get a job. The alternative is saying - "some people die from lack of work, get over it".
I have nothing special for or against the mag, but I will point out that economists are famously known for disagreeing on huge things once you get beyond the law of supply and demand. Not every discipline of study has whole alternative schools of thought the way they do. Economists are closer to philosophists than most philosophists.
I really love economics! But I have no illusions at all that they serve up one and only one version of the truth like physicists. :D
> real, serious and long-standing economic changes
"founded in 1977 by Ed Crane, Murray Rothbard, and Charles Koch, chairman of the board and chief executive officer of Koch Industries."
some of these people have funded climate denial on behalf of the Oil and Gas Industry Lobby.. they have the largest wealth in history to defend, at the cost of Climate Change.
You are aware that disruption has been the human condition for 1000's of years, yes?
Why do you posit that life expectancy will continue to fall? Not sure it's rational to expect an extended reversal of the trends which have been on the increase prior to the past 2 years. Look at the charts and the analyses.
Precarious work? You may need to explain that one.
> You are aware that disruption has been the human condition for 1000's of years, yes?
So what? Slavery and The Plague and Cancer was part of the human condition too.
And its wrong - technological disruption only appreared recently. For ten thousand years my ancestors lived nomadic lives on the eurasian plains, their lives were disrupted when a gun and a steam engine were invented few hundred years ago.
And now we don't die anymore because of simple infection or other preventable diseases. Many if not all of the things that keep us fed, warm and healthy today are only because we endured technological disruption.
So which is it, 'human condition includes disruption, there is nothing wrong with human conditions and therefore there is nothing wrong with disruption
or
'human condition sucks and the only way to fix it is with disruption'?
Are you just making it up as you go along? Do you feel the need to justify disruption at any cost?
There is no need to be angry. No, I'm not making it up as I go, I'm saying that the human condition in the long term significantly benefits from technological disruption, even if that's not always true in the short-term. I'm not justifying anything either, I'm just telling you what happened.
Now, we cannot suppress technological disruption either, we can regulate and channel it, but you won't be getting your pastoral society back in which everything stays as is for a few thousand years . I think that's good, you Kay think that's bad, but it just is.
I am not angry, and I am not even saying you are wrong.
I am just pointing out that the post by halr9000 and your post both try to put disruption on the pedestal, but the arguments are mutually exclusive, and in fact they profess opposite values.
I got a confused, through both posts were made by the same person, hence 'making it up as you go along', sorry about that.
I think art communities should have the ability to set parameters for what they accept. Whether it be something like "only African artists" or "only Holga cameras." If for this one, only generated by people, so be it. If someone wants to set up a community that only accepts AI-generated images, then that's okay too.
Many of the discussions today sound similar to the ones I read about during the advent of earlier creative leaps – which I think merely serves to highlight the importance of this advancement. However, I do still remember plenty of the discussions around DTP and the computerisation of various creative industries. The themes are similar, while some benefits are lost and many are gained - some people are fixated on what is lost, while forgetting that we're growing the field rather than forcing others to abandon the craft. It is, after all, just another tool.
Which brings me to a topic which seems to never go out of fashion: Purism/Conservatism.
Like most discussions that are centred in conservatism/purism, there's always a thread of hypocrisy, because this type of person will use themselves as the golden benchmark to what is acceptable.
To use an example: a purist painter of earlier times might shun the modern purist painter for the simple act of purchasing pre-made paints. They might bemoan that they didn't make their own pigments and mix up their own paints.
So while this online art community might ban AI work, it merely an action that excludes themselves. AI opens the door to many who otherwise wouldn't have the capability of producing beautiful pieces, just as digital painting did, and the number of those people will be larger than current # of digital painters. They should have just made a subcategory on their website for AI and used it to bring new relevance to their platform.
>Like most discussions that are centred in conservatism/purism, there's always a thread of hypocrisy, because this type of person will use themselves as the golden benchmark to what is acceptable.
> To use an example: a purist painter of earlier times might shun the modern purist painter for the simple act of purchasing pre-made paints. They might bemoan that they didn't make their own pigments and mix up their own paints.
That I think is pretty much why this stupid discussion is still going - people considering their work (or just stuff they enjoy) "art" want other things to not be it to feel better.
Anecdotally I dated an artist for a short time who upon entering my home proceeded to ask me how much I paid for my art and why I would pay for something as awful as what I have.
My art was made by a friend of mine who works in a union factory in Texas. Art is his real joy and I was able to snag some pieces from when we were in college together. It's easy to write off that the pieces are just sentimental and therefore subject to sentimental value, but really, his work is very abstract and each piece (if you can snag them in order) describes a time in his life.
I tried to explain to this person how art to me is the story not told or the story as I see it. More recently I had the opportunity to really appreciate his work when I was on an LSD trip, but that's a story for another time.
In my experience, artists are completely divorced from economics, and the concept of the market, while striving to be a recipient of it - making the oppressive desperation their whole identity
For those formally educated, schools should really focus on the economics more instead of relying on magic
The reason to me is that people that buy contemporary art (stuff from living artists) are already going out of their way to do so, knowing that it is divorced from the concept of investment or stores of value. But for other living artists not to really understand the motives or empathize with the customers, going as far to criticize them for any number of reasons (daring to question the stated price, having a taste at all) really undermines the point of bothering with living artists at all.
> Writing a text prompt is a fair place for that line to be drawn in my opinion.
I read an article[0] on the subject the other day. At some point it said that some art tools are inherently easy to use. Part of the difference between an amateur and an actual artist is in the mastery of such tool.
Think of a phone camera for example. It's extremely easy to take more than decent pictures with it. But it takes a good photographer to make art with it.
Bringing it back to the topic at hand it means that yes, describing what you want to see to this kind of software is easy on a basic level. But you get a picture that looks like any other. Instead, you need to speak the language of the software to obtain a better materialization of your vision, and that requires some sort of mastery of that tool. That would be the difference between an illustration made with stable diffusion and actual art.
What I think you miss is that mastery is really just a proxy value for what we really value.
Suppose I'm an artist, with something I feel I want to say, something I want to share with the world. How can I convince you to listen to me, when there's a thousand other people you could listen to? Especially about big issues like what's worthwhile in life.
Well, one way is commitment. Supposedly, Sophie Germain decided as a child to study mathematics because she read the story about Archimedes, who was so obsessed about his circles that he got killed over it. Clearly, Archimedes had found something that really mattered to him, and if Sophie studied it, maybe she would find that meaning too?
So likewise, as an artist I can convince you to take a look by being really dedicated, practicing the craftsmanship side of what I do so that no one can convincingly say they do it better. That doesn't guarantee that what I say is worthwhile to you, but it's at least an indication it was worthwhile to me.
But you know, the converse is not necessarily true. Someone who wasn't the best at the craft and who didn't sacrifice everything for the opportunity to reach you, could still have something wonderful to say.
Sometimes the skill at the craft limits what you're able to say, even if what you wanted to say was great, but I think that's rarely what matters. On a technical level, there's very little that distinguishes, say, the best guitarists in the world.
I like this train of thought. I wonder if "text in prompt" is far too simplistic of an interface as an input to a black box to consider skill level with?
My inclination is to say the interface needs more breadth of options beyond single-line string input, and the outcome needs to have some level of predictability to enable the artist a significant enough level of understanding of how each aspect of their input alters the output.
I mean, the input parameters should be complex enough to let artist intentionally develop a unique style/idea, rather than style being dictated inside the black box.
If you used an AI to write a million DALL-E prompts for a million different outputs - are you an artist if you pick the nicest looking output and call it art?
Well it’s a human pointing the camera and pushing the button. It totally depends upon the camera how much influence it has on the actual picture produced. Depending on your settings, hardware control and image processing can be applied to automate the production of a more acceptable image. Is the camera or the person holding it producing the image? Or is the photogenic environment the camera is being pointed at responsible for the resulting image?
Once I’ve selected one from the 20 or so images, had it printed, framed and placed it on my wall then weeks later notice it again when the light coming through the window catches it just right, I think that is a really great picture that transports me to another place.
Is it art? Because I was holding the camera does that make me an artist?
All I know is that my life is enhanced by having that image on my wall however it came to be there.
I'll always remember 2022 as the year that creativity died.
That sounds a bit negative I guess, but techniques like stable diffusion will be applied to other problem domains, so all other art forms like music and creative writing will be automated within the next 5-10 years. And if computers can do that, they can do anything. Because that's how our brains work. Our neurons are microscopic filters magnifying patterns in the long tail of quantum randomness where source consciousness probably resides. Although science will only have recipes for creating consciousness, it will never be able to explain meaning, because it has nothing to say about it. That's the domain of personal experience, faith, the study of magic, etc.
The big question for me is: if a liberal arts degree is worth even less than it is today, what will people do for work? We're looking at not just the death of creativity, but the death of all human contribution. This was predicted decades ago (see Asimov, Kurzweil, etc), but we're not prepared for what's coming. I just don't see us switching to UBI before we're subjugated beneath the yoke of corporate AI that gradually changes out human workers for machines. We're in the endgame now.
I still hold out some small hope that kids today will run with this and do creative stuff we can't even imagine. Or will we just merge with AI when the Singularity arrives around 2040? Who knows.
As a hobbyist game developer, 2022 is the year my creativity can finally fully unleashed.
Prototyping games is getting so much better, this will really help indies game studies quite a bit.
Will those AI generated assets replace hand crafted ones? Heck no!
Have you ever tried doing AI art?
What gets posted on social media are the few successful experiments. Most AI generated stuff is absolute crap. Getting anything useful out of it requires lot's of trial an error and magical incantations.
Now yes, you can quickly make generic background #20233 with it or simply something weird that would fit into some bad dream. If you need something specific and coherent though? Yeah, this will be very hard.
Look how many iteration that requires. How it requires finding the proper real word assets for img2img techniques. How is still requires some serious Photoshop skill to put everything together.
AI helps us build more higher level tools for creating art but the idea that it is anywhere close to creating art itself is silly. We have not made any progress in creating general artificial intelligence at all.
The main progress we see is that AI-tech is that it is becoming available for the masses, which is awesome.
> What gets posted on social media are the few successful experiments. Most AI generated stuff is absolute crap. Getting anything useful out of it requires lot's of trial an error and magical incantations.
You are exaggerating. Easily over half the output I generate is on par with the standard being posted on social media.
> Now yes, you can quickly make generic background #20233 with it or simply something weird that would fit into some bad dream. If you need something specific and coherent though? Yeah, this will be very hard.
The demos look amazing because there is no acceptance criteria. Anything gets hailed as "groundbreaking" because is process is novel. Which is a completely different scenario from what you expect from someone actually being paid for making art.
I am still figuring that out myself. My current project is a fantasy-themed visual novel for which creating backgrounds is super easy. You just pick a certain style, for example "by Greg Rutkowski" and put that in any prompt.
Normally you want to reuse your backgrounds as much as possible to not blow the budget but thanks to AI I can be much freer with my writing. Of course the important backgrounds should be remade by hand at some point but for prototyping they are great. Character art does not really work beyond concept art as you normally need different poses for a character and you are very specific about what you want, again not a strength of AI.
My advise would be to play to its strength, if you can: Horror works really well. Outdoor scenes work better than indoor.
I think things will categorically change about how we define art in the next little while. We may see an increase of parasocial-style livestream art drawing, so the illustrator isn't just an illustrator but specifically a personality attached (an exaggeration of the current artist-art relationship commonly known today). We may also determine AI-generated art is simply in another category, like photography, CGI, digital, and traditional portraits are all considered different categories. I can imagine human-generated novels and AI-generated novels are neatly separated, with associated scandals if a human attempts to pass an AI-generated novel as their own creation (scandals already currently exists for humans who eg. a romance writer recently got in trouble for lifting paragraphs from fanfic they didn't write into their original fiction).
There's nothing to say that creativity will die. We invented typewriters, but calligraphy is still an art whose pens and inks are beloved interests. We invented the digital screen and the paper machines, but handmade paper and hand-stitched book binding is still an artist's pursuit.
I dunno. Fast food haven't killed restaurants, and restaurants haven't killed home cooking. Fast food is a huge problem, but many go to restaurants and many still cook at home. We're in no endgame at all. If it worked this way, artists from poorer economies would have drowned out all other art. But for some reason, it just doesn't work that way.
There's a few artists currently doing creative things with stable diffusion that aren't just "put in a prompt and let SD crap out an image". I think this music video from Holly Herndon shows the promise of this stuff. I do think ultimately once people have been exposed to enough AI art they're going to demand more creativity. Right now there's a lot of exploration going on and people need to get more familiar with AI tools before they start doing amazing things with them.
>I'll always remember 2022 as the year that creativity died.
It's like saying that computers killed artists. AI is just a tool, it still need a human in control and guiding it, can't generate anything on his own (yet). If anything, this can help creativity.
If a computer is a bicycle for the mind, then an AI is a mind for the mind.
Admittedly, my acquired lack of creativity after a lifetime of struggle is the very thing preventing me from seeing beyond this horizon. Too old to begin the training and all that! Ironically, I find the not-knowing to be somewhat comforting.
Exactly, think of it as a symbiosis, a man is slow, but a man in a bycicle is fast. Now a man in a rocket, can go to other planets. But it is still a man, a rocket is just an inert pile of metal and fuel.
Same with AI, AI is the rocket, it propels and accelerate creativity to inhuman levels, but there is still a man controlling it. AI is just an extension of our minds like a rocket is an extension of our feet.
Yes, and just like in communities described in this article, AI is banned from human competition. You can use it to practice, but you can't go to the tournament with your own chess engine and say that you deserve that win, because you programmed this chess engine.
Yes, but that's because you know when deep blue is playing chess. The recent accusations that are rocking the chess world are that one of the champions may be cheating. If we get to a point where cheating can't be detected, that may do quite a lot of damage to the game.
I wouldn't worry too much about AI stamping out creativity, all the results I have seen just look like the type of art you see in hotel rooms and hallways. It's supremely ignorable stuff, possibly exactly because the models are built from billions of samples and by definition most of those samples will be average and below.
I will never understand you AI optimists.
AI will never create remarkable or even widely popular music. Well, never say never. But 5 to 10 years? Not a snowballs chance in hell.
> Our neurons are microscopic filters magnifying patterns in the long tail of quantum randomness where source consciousness probably resides
gibberish
> Although science will only have recipes for creating consciousness, it will never be able to explain meaning, because it has nothing to say about it. That's the domain of personal experience, faith, the study of magic, etc.
misguided conjecture
> We're looking at not just the death of creativity, but the death of all human contribution
hilariously misinformed extrapolation
AI is nowhere closer to cracking the meaning problem. Until we have meaningful AI generated art there's no point for this baseless speculation.
I paint a fair bit. Artists all know how hard it is to innovate, and how difficult it is to faithfully physically represent ideas, images, and emotions that they experience internally. People don't respect works created by algorithms because people feel algorithms (1) do not have a rich internal state, that is (2) challenging to express, (3) will one day die, and (4) never occur again.
There's analogues in music. We have algorithms that can write jazz standards with all the same essential characteristics of Blue in Green, or So What, but no one cares. People just want to hear Miles Davis and Bill Evans do it.
> We have algorithms that can write jazz standards with all the same essential characteristics of Blue in Green, or So What, but no one cares.
That's definitely not true. In the areas that make money with some semblance of reliability for the artists (low budget film / game scores esp. asset flippers, shitty TV shows, music for ads, etc), people are going to be more than happy with a magic "make music" button without having to pay for a composer. Like, if you thing that anything in this video is out of reach for today's AIs... https://youtu.be/QgU9H9dECt8
I agree 100% with this. The only thing I would add is that a key property of art is that it must make the audience feel something. It seems very clear that AI-generated art is making people feel something, and I think that what it makes people feel will change over time, as these tools become more accepted in the art world.
I also think it's very interesting that these tools enable a near absence of technique. I tend to be one of those who doesn't value technique very much. For instance, I believe I've commented on this site before that while I certainly appreciate the time, effort, and technique that goes into photorealistic painting, I don't think it brings a whole lot to the table beyond what a photograph would. This is strictly a personal assessment, of course -- others are free to, and do disagree with me. I just feel as though photorealistic art leaves less room in it for the viewer than other forms. (Incidentally, I think this also applies to AI art, and may be a part of the reaction to it.)
I totally agree [1]. Once all the panic dies down, I think you'll see these technologies enabling artists more than hurting them.
The displacement problem isn't quite there yet. While it's a good replacement for stock photo aggregators, it's not a great replacement for commissioning an actual artist. E.g. if you want a consistent look and feel for each of your images, you'll struggle. It's only the C-listers, who create heaps of bland stock photos in hopes of earning royalties, who have something to worry about.
As the technology improves, you'll see a whole new tranche of creators who are able to harness it to create amazing new things.
> Once all the panic dies down, I think you'll see these technologies enabling artists more than hurting them.
This has happened in a different (non-AI, but "technology enables") way already with other forms of creativity. While it is true that more people become enabled, its not clear that this is overall a good thing. The benefits of more people empowered and enabled to engage in self-expression has to be traded off against the problems associated with discovery caused by their work becoming a part of the cultural milieu. People making art in private is almost certainly a good thing without limit; people making art and then making it public is not so clear - most of the work people create (even the best) is just not worth being public, but the self-limiting discrimination that this implies is in short supply in most of us. Yes, it's good you made it. No, it shouldn't be online or public, almost all the time.
I think it'll displace a chunk of what artists do. Anything where there don't have to be super precise specifications for the end result, where there's some leeway and you just need something that goes into a general direction. Which is a non insignificant chunk, and maybe kind of unfortunately also the jobs where you can be more creative about what you're doing. Like if I'm making a magic-like card game and I need illustrations for my playing cards, I could use an AI for that. If I want a comic strip or an illustration of a specific object, it's probably a poor fit.
Same here. Software dev with an MFA in Studio Art. Generative Art was a topic 20 years ago in school and my feeling it was politely tolerated as conceptual, but not aesthetic. The difference now is that the models and algorithms and amount of data actually produce decent results. GPT-3 writing as opposed to the markov chain experimental writing decades ago is profound. Image creation even more so.
I see exactly this backlash in most art schools until eveything is so mingled a new generation just doesn't care.
I like the more eastern idea that art is mastery of skill and art pieces are more of a byproduct we can observe.
Photoshopping or making an AI make art still takes skill, so it's still art - just in a different category. Putting a robot in a boxing match or an AI in a human car race wouldn't make much sense, but robot vs robot would.
There is a word for that in the west as well, we just call it craftsmanship. The artist is the one we allow to decide what the purpose of the work is, even if they outsource all the work to apprentices, assistants, or for that matter computers.
This is the kind of comment about art that is measured, constructive, non-absolutist, and too rare on HN in posts about art (and urban dictionary has an entry for 'reverse monet' cos I wondered, but I'd skip that one if I were you).
aside from showcasing, you have to factor in how the bottom is now out from MANY freelance artists who were already fairly lucky to have clients paying them $5/pop
for example, in the last two weeks, promotional flyers for parties in my area have been all AI generated, because Stable Diffusion is free and DALL-E 2 completely fumbled with their limited access and data credit system
its kind of game-over for the promo art trade, the transactions that were happening, are not happening
>Back when I was in school, there was already plenty of debate about 'what is "Art"'? It's always been hotly contested. Most of my teachers felt that 'technique' was not Art. Others were more purists – life painters, etc – and felt that, no, technique is art.
Of course people practising it would say that it is an art.
And of course they would discount stuff that is beneath them as not art. "I'm creative, I don't just copy stuff I see!", just to feel better about themselves.
So someone might say code is not an art, it's same as any other engineering. But code can clearly create art pieces, whether directly (say a shader generating pretty picture) or indirectly (AI training and generating pictures based on that).
So, what, do we just say "art is any product that you can experience and have feelings about?". Nah, can't discriminate this and that based on that vague expression! We might even accidentally call blacksmith an artist if we go too far! /s
> This is a really interesting debate – I just wish it scratched deeper than the surface more often than it does, with a bit more historical context.
I find this is true of a lot of topics, too bad there's not some way to shorten the time between shallow to more useful/informative debate. Your post above does that for this topic, thanks for that.
Others have chimed in already but I was inspired by this idea:
> Like a music synthesizer times a million.
Even now, every once in a while I find people arguing that synths are not "real music". Also DJs, etc. generating songs using pre-made samples. I think it's a very apt comparison, where tools become more and more available and barrier of entry gets lower, I don't believe art will die but rather just change into new forms, and what's now revolutionary will become just a very commonplace tool in the future, just like synths, and "real" artists will keep on existing alongside.
I suppose the 'photocopier' argument is pretty similar to collage - sure, it's recombining existing things, but presumably they'd all agree Max Ernst wasn't just a machine?
I enjoyed this work called 'art is' by poet George Quasha that consists of interviews with artist answering what 'art is'. hundreds of answers over decades and no two the same
https://vimeo.com/323912060
Typing prompts into a program until you get an image you like does not make you an artist.
Duchamp was already one of the great painters in the world when he showed the readymades - and there's a lot of evidence that he made them, what a great joke if so.
Aside from the debates about "artistic merit" and the ethics of using AI trained on existing artists' work, one of the key things is that this new technology allows anyone to flood a place with images.
There are some scifi-themed subreddits that I visit that allow posting art, and at this point it feels like 50% of all submissions are Midjourney/Dall-E generated art. Fine, it looks ok, but there's such a low barrier to creating and posting these, it's beginning to drown out other valuable content. The article even speaks about this:
> In July, one AI furry porn generator was uploading one image every 40 seconds
>Aside from the debates about "artistic merit" and the ethics of using AI trained on existing artists' work, one of the key things is that this new technology allows anyone to flood a place with images.
Just subbing to the StableDiffusion sub has added a ton of art to my feed in Apollo. I agree the concern isn't that it's art, but that it's being produced so fast. I've seen dozens of showcase/portfolio worthy pieces generated in the past week. That manually created would probably take 40-160 hours (I'm not an artist so hence the huge range).
It's pretty comparable to handcrafting a letter vs. a spambot in terms of the volume jump.
This could be solved on basically all arti communities by simply putting in a post limit per day. Say an artist can only upload 5 images per day or something, why would someone need to upload more than that?
I mean, practically it means that you haven't solved the problem of "now the website is flooded with content". Arguably, the core problem is just that the premise of the website fundamentally doesn't scale to having too much of a maybe-good thing.
More like one image per X days. I don’t care how many different things you can type into an AI in a given period of time. I only want to see the best of the best of the best.
This is like when we discovered stereo recording, and mixers would put tracks at either 100% left or 100% right, just to show off the novelty of it.
Or when 3D movies became ubiquitous and every movie (even those shot completely without the third dimension as a consideration) got released in 3D.
Once the novelty wears off, I imagine things will be better.
For now, it’s just a deluge of low-effort, yet high (relative to traditionally produced art) apparent quality artworks.
> Say an artist can only upload 5 images per day or something, why would someone need to upload more than that?
There's been more than a few days when I've uploaded more than 5 images a day to flickr/Instagram (according to my backup, I uploaded 27 images to Instagram on 2007-02-04 during a day-long walk.)
I think this will subside significantly once the novelty wears off, and examples of truly outstanding generated art become more prominent (establishing a consensus baseline for quality).
It's important to keep in mind that these art communities represent one very narrow kind of art, one that is more focused on illustration and graphical style, and most often not the kind of art that major art institutions work with. In the past 5 years, AI-generated art has been accepted and incorporated in the contemporary art world, with major works produced by luminaries such as Pierre Huyghe (of Ideal), Trevor Paglen (Bloom), and Hito Steyerl (Power Plants). For these artists, the graphical potency of the generated images is usually only one factor of the work, which also centers on how the artists use the technology, its origins, history, and broader uses, and what it might imply.
For this kind of artwork, the kind made by artists who feel threatened by AI, I’ve been thinking for a while that a computer could likely do it better. Intricate detail, repetition, sampling and recombination of popular styles, optimization for aesthetic popularity and mass appeal — these practices are all squarely in the domain of automation, data science, and marketing. We might call it the Netflix-ification of illustration art, but it’s quite different, because the tools for making are available to many. As an arts educator and artist, I welcome these new tools, as they allow easy experimentation for newcomers, but also as they reveal that art based primarily in deployment of technical skill and eye-popping graphics is missing something vital.
For artists who continue to paint, or use other physical means of expression through base materials and sell them as physical works, this entire AI art surge remains largely irrelevant in a practical sense. A buyer or patron who wants real, physical art will still need and want real, physical art, and I don't see an AI imitating much of that easily any time soon.
The people or organizations to which AI-churned art appeals to most desire commercial visual churn. They are the same people who would in any case have least cared about art in its more creative, communicative forms. And there's nothing creative or imaginative about the results of machine-generated visuals. They're the very opposite in fact, visuals created literally mindlessly.
If anything, I can see AI art giving a new breath of life and interest to real man-made art, much like industrial food production gave organic "local-grown" food producers a chance to brand themselves with a distinct, desirable identity.
I don't know if I agree. One of the big sources of income for online artists is commissions of D&D characters and such; AI-generated art is now getting good enough that people can get a "good-enough" version of what they want for free, instead of paying an artist $50 or $100 for a piece of art.
I've commissioned art for my D&D character before, and I'll almost certainly do so again - but if I'm playing a one-shot online and want a quick image to share with the group, AI-generated art is a fast and free way of getting that!
People have been losing their jobs because of technology ever since the industrial revolution, each one of them enraged about it, and rightfully so. Society doesn't care about people who've put their whole lives into a skill that suddenly became worthless (or much less valuable then it was).
The solution is not to ban technology (you can't really ban technology, as it's not under total government control) - the solution is to enable anyone to pursue new skills without fear of homelessness and starvation. I.e. Universal Basic Income.
i mean what is even the point of pursuing a new skill if you can't create things anymore because the act of creation and craftsmanship have been so thoroughly devalued in favor of cheap volume? what kind of fucking joyless bugworld is that?
Plenty of craftsman have good careers. We have a cabinet maker in town that is busy charging double the other cabinet makers. Those other cabinet makers charge double what a crew will install from Home Depot. All of them are busy.
>The people or organizations to which AI-churned art appeals to most desire commercial visual churn. They are the same people who would in any case have least cared about art in its more creative, communicative forms.
I specifically addressed this in my original post. You're exactly the kind of person who would possibly like this AI technology, and apparently not someone who would have an appreciation for art in its artistic, emotionally appealing form. Your disagreement is based on a misunderstanding of definitions.
All of you commenting seem to miss the fundamental point that I made: People who appreciate art for art's human aspects will want HUMAN, emotionally derived art, not a printable piece of AI-generated pixel-vomit that can then be mounted and varnished. I have nothing against anyone doing this and I absolutely don't support banning any technology that lets people explore new frontiers of creativity, but I don't see even AI-connected devices that can physically paint things replacing artists as imaginative human creators of concepts and emotional ideas expressed for other humans. To go on about how an AI could eventually paint or sculpt and so forth from its own algorithms is to miss the point entirely about this specific type of creation and the extremely old human desire to enjoy it.
Perhaps to be expected from a site so frequented by many people who spend their working lives "optimizing" X or Y gimmick for major tech giants and viewing other humans largely as simple parts of a data gathering landscape.
>For artists who continue to paint, or use other physical means of expression through base materials and sell them as physical works, this entire AI art surge remains largely irrelevant in a practical sense. A buyer or patron who wants real, physical art will still need and want real, physical art, and I don't see an AI imitating much of that easily any time soon.
Giclée prints on canvas are already very popular with artists. Printing into physical medium is already trivial whether the art is produced by human or AI.
See as an example the Epson SureColor S80600 Printer:
For artists at high levels who print, the print is a part of the work and is often done with close to as much care as one puts into a painting—fine tuning colors, making small details perfect, printing on exotic materials, totally messing up the print in certain ways, painting over it, layering the print with other things, printing on fabric and hanging, stretching, modifying, or destroying the fabric etc etc.
Every aspect of making things available to humans becomes a medium, from writing code to literally just cleaning something to factory production to using "art materials."
Indeed a lot of mid-tier / long-tail work is also just straight up prints from the print shop, but OP is talking about the bespoke side of the spectrum. At the high end, fine art defines "bespoke."
And then what is the story behind the work? Well: an artist made someone with DALL-E, printed it, and then printed over it to make it convincing.
Now, compare with: "this is an original master work of a historically revered artist," just to take _an_ example.
Alternatively, "my friend made this by painstakingly carving into a 200 year old block of hardwood taken from the Black Forest after the death of their father."
Certainly you can fool people, but what's the point of that?
Not to nock it, but things are what they are. Art is received as the entire experience of the object. Often a simple print is the right thing for the context, but that's not what OP is getting at.
The point is it looks nice enough for gedy to hang up on the walls. I presume they don't care about fooling anyone, and likely would readily tell those who ask that it's an AI image. Sometimes people value the aesthetic over the provenance or story behind a piece of art.
> For artists who continue to paint, or use other physical means of expression through base materials and sell them as physical works, this entire AI art surge remains largely irrelevant in a practical sense. A buyer or patron who wants real, physical art will still need and want real, physical art, and I don't see an AI imitating much of that easily any time soon.
I don't think it would be that hard to get a machine to physically paint in layers. Sure, the velocity, pressure, texture etc. of a brush would take time to model, but you could start with spray paint.
Importantly, a physical painting machine isn't gonna be something anyone with a beefy GPU can download for free. It would pretty much only be useful for factories in Mexico making "Henry Rogers"-style mass-produced paintings [0], which would be bad news for the impoverished workers currently making them, but not for most other painters.
Yeah maybe the demand for visual art just isn't there. I imagine a plotter with a brush which has paint injected directly into it, which I don't think would cost that much materials-wise.
I was looking for this reply. Imagine if you needed to hire an artist but 9/10 listings are from alibaba express, full of fake reviews (other nonartist techies commenting on ai art)
Or maybe because this is where artists place their portfolios and theres no reason to make ot that much harder to find.
It's not indistinguishable, it's just overwhelmed by volume.
This is like giving someone the needle-in-a-haystack problem and asking "Why do you need the needle when it's indistinguishable from this easily-findable piece of hay?" The hay isn't useful to me, but now you've given me all this hay to sift through before I can find my needle.
Then why do they pretend to ban a specific art type?
I could have generated variations of a human created digital piece in less time than these AI (currently) since the creation of this platform but they didn't seem to feel threatened by this up until now... so there is clearly an other problem.
The point I was making is that prior to AI generated pieces, a human could create one piece and then apply random filters to it and also upload thousands of small variations every day. People didn't do it because it is not very interesting. The broader point is that these AI pieces can be interesting, and such a ban is not just because of bandwidth/storage limitations.
I generated over 30k, maybe 40k images since stable diffusion 1.4 model went public 2 weeks ago. Most of it is just endless exploring and generating a lot of nonsense but I got some prompts that generate consistently good results. 2000 decent quality images per day would be quite easy, even with my humble rtx 2070s GPU.
It's interesting to think about it, what differentiate art from a human or a machine. It can't be the content itself, because now machine output is close to a human output (modulo the small details).
So what make those communities attracted to art made by humans, rather than art made by machines? It must be the "act of community" itself, which cannot be made by a machine (it is useless to be "in community" with a conscious-less machine).
So this would mean that the "art content" for humans is not only the pictorial content of the image, but also the "transmission of a message", and in particular, the quality of the initiator and the receiver of the message. In this case, humans would (and apparently seem to) differentiate in quality art that has been created by a conscious being and for a conscious being, to art that has been created by a machine.
Has nothing to do with that. It's two things. Artists are threatened, and also right now a lot that is being submitted is low quality.
So if you are an artist you don't like seeing a bunch of AI junk on your home page just because a lot is junk but also, because one can assume that it will continue to improve so that that most of it really is competing with artists.
So that could put a lot of people out of work since it would make art-related business models much more competitive.
> So this would mean that the "art content" for humans is not only the pictorial content of the image, but also the "transmission of a message", and in particular, the quality of the initiator and the receiver of the message
Sounds like you've taken a semiotics class.
If art is about transmission of a message, wouldn't the message be the prompt supplied to the AI by the human?
In that case, perhaps the art is only as profound as the prompt?
People consume art for different reasons. It can be like a spiritual experience, or an investment, or something to brighten up your life. It can have a story attached to it or be inspirational. My guess is that different people will have different reactions to machine made art. And most of the art community will carry on perfectly happily as if nothing has changed.
Individual art communities should be allowed to decide to only accept art created by a particular set of tools. Oil. Acrylic. Mixed Media. Watercolor. Chalk. Pencil. Digital. Photoshop. AI. And so forth.
It's a matter of curation in my mind, not discrimination, and not really outrage worthy.
I didn't take the idea that this was shared as an outrage piece as much as a point of interest; it's a key point along this journey and worth noting with or without judgement.
I agree that images only through AI should be removed from art sites.
The issue with AI art is that the only human contribution in those pieces is the creation of the prompt and the curation of the final image. Without sensory and experiential inputs of its own, AI can only interpolate between real images and styles. So an AI generated image doesn't add much artistic value since its not expressing any new deep truth or factual understanding of the real world.
Well isn't all art just a derivative of another piece? The issue with that statement is that art is fundamentally a communication from a human artist to a human viewer. At the level of AI prompt creation and curation, you act more like an art gallery owner rather than an artist, which doesn't communicate artistic values in the same way as does an original piece derived from the experiences, history, and skill of a human artist.
Dall-E 2, Stable Diffusion, should be used for inspiration and helping with minor elements of a piece, but someone passing off an AI generated image as part of their portfolio is disingenuous at best
"Truth" and "understanding" are both things that have changed throughout human history, they are not fixed. Art can be something that changes the way you understand the world, which would change your idea of what the truth is about reality.
I just rewatched Goodfellas the other day. No doubt that film brought a deeper understanding for many many people about mafia philosophy, history and events, and inspired more research (I went down a Wikipedia rabbit hole after watching it, myself). Dali's melting clocks might suggest a new way to conceive of time. I only dabble a little in science fiction, but my impression is that it has historically inspired plenty of scientific and engineering endeavors, including AI generated art (and I'm sure there are plenty of dystopians out there that may show how it could go wrong). Reading literature from everything between the ancients and contemporaries, I'm constantly finding things that parallel my current interpersonal challenges and give me new insights.
Your post is just a typical arts-based response. Not rigorous, sloppy, with "No doubt", "might suggest", "my impression is", "interpersonal challenges". It's just a sloppy mess that can't be engaged with, quite deliberately.
Edit, if I sound frustrated I am. I'm trying to get a better handle on 'art' so I can understand what is there and what its genuinely worth to other people. It's rare to get anything other than gloop, from which I'm starting to wonder at the bona fides of people who say supportive things of art. Generative art is frankly amazing - how much input was from artists (any at all?), how much was from scientists/mathematicians? (and how much from the intersection of the two if that's not empty).
> "Truth is the property of being in accord with fact or reality"
I asked what is truth to you, and you regurgitate something from Wikipedia with no synthesis of your own. OK, now why don't you copy and paste the definitions for "fact" and "reality", and make sure they don't contain any synonyms for "true" or "real". We can keep iterating from there until one of us passes out from sheer boredom.
> can't be engaged with
Nonsense. You could, for instance, engage in showing me just how non-rigorous my statements are. You can start by proving that every single person who has ever seen Goodfellas came away from it with zero delta in their understanding of the Mafia or inspiration to learn about it. Including me. And then prove to me whether their understanding was "true" or not.
And then prove to me that I didn't think of new ways to deal with the problems I had between myself and some friends as I was reading Proust. And/or that I would've thought of those things if I never read Proust.
And then prove to me that science fiction has never inspired any scientist or mathematician in the history of science fiction.
And that there has never been a person that looked at Dali's melting clocks and wondered if time is more complicated than what they currently thought.
The problem is that it can take a long time to explain certain feelings or ideas you get from art (of any form), and even the attempt is a bit painful or downright impossible.
A very low-level example would be a Jackson Pollock or Mondrian painting. It's just a bunch of... stuff but it's extremely visually pleasing compared to just random blotches. There are theories as to why that might be[0] but overall, it would be nigh impossible to tell you why I personally find these pieces compelling because the reason is almost unconscious and tactile.
A medium level example would be a song that inspires a certain narrative in your mind. For instance, that intro song by the XX gives me all sorts of conflicting emotions that could be summarized as 'wanting to do more with my life', but the summary doesn't do it justice at all.
A higher-level example would be a piece of literature read at the right time in your life. It's a show-don't-tell type of exercise where the author exposes ideas to you about life or the world without spelling them out explicitly. Sometimes the ideas don't have much to do with the surface details of the narrative. In many ways, it is similar to experiences you actually lived yourself and that imparted ideas or lessons to you.
If you are not in the same headspace, this will seem like a con or pretentious nonsense. But what is really going on is that trying to explain it is often like trying to explain a joke or a subtle social interaction to someone who didn't get it.
My friend once tried to explain to me the beauty of the Carmack fast invert square root, and I didn't really get it because I lacked certain foundational ideas. But there was no doubt in my mind that he was seeing artistic worth in that moment, and that despite me being unable to understand it it was certainly not 'gloop'.
Trying to come at it with the combative Wikipedia-article-on-fallacies/STEM vs. art mindset is self-defeating, like interpreting the scientific method as a tool for prescriptive claims.
I don't have time to do this very good answer justice, I'm sorry. Let me try and be brief. Some people like Pollocks or Mondrians, some don't, and one thing I'm asking really is that people be honest and not pretend. Such honesty is a kind of rigour.
if we accept that certain kinds of art are liked by some people and not by others then questions have to be asked about what I some art is considered good generally (and goes in the gallery) and some isn't. To ask this question is also a kind of rigour.
Art I don't believe can be justified in any scientific sense, but I think the art may be judgeable in its own context, and I'm fine with that, provided we try. This too is a kind of rigour.
I perfectly understand the your fast inverse square root, or Euler's e/i/pi/-1 formuala are beautiful to some, they are not to you and me. Let's be careful with that, try and understand, and tolerate people's differences instead of saying "this is beautiful" and "this is not beautiful" (because that's an absolutist position). This is a kind of rigour.
I agree, scientific method is not appropriate, but something else is. Let's try to find it?
Anyway, that is an absolutely excellent answer and while it doesn't answer my questions, it does open up the hope that these questions can be answered or at least approached. For which, thank you!
Edit: somehow some just is bad, check out http://museumofbadart.org/collections/ I'd wear dark shades if I were you, some of this stuff blisters the eyes.
There are bound to be philosophers who have tackled these questions in detail. I can't make any recommendations here as I lack the requisite knowledge, but I do have a proposition that starts with the following question:
"What sort of knowledge and information do we rely on independently of the scientific method?"
We can't come up with an exhaustive answer, but what immediately comes to mind would be social skills, gut feelings, "common sense", familiarity with discursive and rhetorical styles, any understanding of the world that has no control group and can probably never benefit from one. Of course, there are plenty of counters to any of these elements. You could say that some people deploy the scientific method in some way with social skills and so on. Or point out the LessWrong community's daily use of Bayes' theorem as they live their lives. Overall, the vast majority of our daily decisions are non-scientific and driven by feelings and intuitions, even if you can apply a top-down conscious modification of your lifestyle to meet goals, better your conditions over the long term etc.
The main conclusion we are left with is that there is a plurality of understanding across human beings that is not quantifiable but decently predictable across an undefined range of subjects. You can make a joke in a certain setting, and expect people to laugh fairly reliably (assuming you are good at this of course). You can talk to HN users and have a strong intuitive sense of what sort of discussion they might want to engage in. A person who is naturally talented at humor will not be rigorous in an explicit sense, but their ability to be funny will be rigorous in its rate of success. That bad art website you linked to (correctly) assumes that there is a sense of what bad art is and that much of the population shares that natural-mixed-with-cultural intuition. They publish the website with the expectation of the audience understanding it.
So essentially, the 'something else' we are looking for can simply be to accept that when we talk about art, we talk about a predictable understanding that is self-hosted across multiple sentient serv... uh, sorry, minds we are likely to interact with, and that can also be reliably understood from scratch. Now, you probably let out a groan at this point because we seem to be essentially back to your initial frustration: a bunch of people whose intentions aren't clear talking about equally unclear things as if they were obvious to their listeners, and being slippery when questioned.
But! the key addition here is the "understandable from scratch". You can expect any human being to understand artistic value to some extent (unlike, say, Terence Tao level math which I am confident I'll never understand), but gaining that understanding will require social and cultural engagement over a period of time on their part. The rigour we seek is something we will have to demonstrate ourselves in the intensity of our efforts to engage with art and its context.
One reason that good art is often defined by its ability to stand the test of time, is because this indicates a solid reliability in getting people across very different eras and mental worlds to engage with it successfully and be willing to 'steelman' the artists' vision.
> The issue with AI art is that the only human contribution in those pieces is the creation of the prompt and the curation of the final image. Without sensory and experiential inputs of its own, AI can only interpolate between real images and styles. So an AI generated image doesn't add much artistic value since its not expressing any new deep truth or factual understanding of the real world.
That kind of reasoning would throw entirety of Realism into "not art", because it is also not trying to express any artistic value. And there were arguments back then too
I guess whole debate about Realism could be just copy-pasted to discussion about AI arts...
If you just want a beautiful or intriguing image to look at, an AI-generated image may happen to satisfy that need very well.
The same need can be satisfied with a photo of some natural things, e.g. living beings or minerals or places.
In both cases there is no human to whom one could attribute any artistic merit for the content of the images.
Whether an image is beautiful or not is not necessarily correlated with the involvement of a human in its creation, but its origin should nevertheless always be made clear.
>In both cases there is no human to whom one could attribute any artistic merit for the content of the images.
I'm not sure this is fair. How you create your prompt at the start is going to have widely varying results on all sorts of aspects that are choices made by human artists that impact the quality of the final piece. Then there are a variety of methods, including img2img, inpainting and outpainting, etc. In all of these the AI is doing most of the work, inasmuch as putting any pixel in any place, but it's not making all of the decisions.
And to be clear, I'm not some sort of AI maximalist - I have a large enough collection of human made artwork that a lot of it is stored in flat files because it wouldn't fit on my walls. I'm not planning on changing that any time soon, and will continue supporting artists I follow.
There's a massive flood of crap out there. But some people are consistently making things that look quite nice that aren't just "thing x in setting y, 8k, artstation, by wlop" - there is real thought put into composition, etc.
I don't know how much artistic talent this takes - I've spent a couple dozen hours playing with these tools over the past week or two, and I cannot reach the levels these people are - but it's more than none.
So "AI" art isn't ok. But presumably Photoshop as a tool to create artworks is ok? What about some of the filters and plugins available?
What about if you use AI as a tool to prepare art that you then make yourself? I was looking at SD and thinking this - I make linocuts and have been experimenting with using it to create layer ideas that I then cut by hand. Is this ok?
What about using AI purely as a prompt? I tried just "linocut in 3 colours of a seascape" and got a whole nice bunch of starting ideas. Is this ok?
Seems to be that there's a huge amount of nuance here that's pretty fascinating to watch / talk about.
I think we should also be aware of the technological reasons for separating AI-produced and human-produced images: New effects/risks can occur when the ML system is naively trained against its own outputs. It's not hard to imagine multiple companies trying to periodically scrape and build new image datasets and retrain the newest models against the updated dataset. But if stablediffusion vN is mostly trained against stablediffusion v(N-1) images, then it's possible for the sequence to not actually improve even though more data and computation are being thrown at it.
They should also ban people that have seen more art than others as they may be inadvertently creating derivative artwork from their more thoroughly trained real neural network in their brain.
Purposely being facetious, but seriously, we will likely see more and more of these type of rules being applied to competitions or groups. It will be harder and harder to define the rules. At some point, it becomes moot. Did I draw it? Did my toddler? Was I on hallucinogens when I drew it? Did I use computer algorithms (filters) on any form of the digital art? If the intent is to preserve the "effort" that goes into art, fine, but that will be a really gray line. I guess it's all fine to have the rules, it's similar to the natural body building competition.
Newgrounds: "not have the Art Portal flooded with computer-generated art"
DeviantArt: "complaining that their feeds are getting flooded"
Artstation: "the front pages are flooded with unmodified AI generated slop"
They're really worried about the volume that AI art scales. So they don't really have to chase down every image and decide if it's AI or not. They just need to stop users from throwing huge amounts of them, which is easy to detect.
I contrast these reasons with the FurAffinity announcement from a week ago[0]:
> AI art is a damn good learning tool, but it's taking everything it does from artists. It leeches on the time, effort, skill, energy an artist put into their work.
> You can use tools like this to learn and try to come up with concepts to learn from. I have nothing wrong with that, but just adding it to an image gallery. The origin are questionable unlike, say, using a site specifically for royalty free images.
How can they draw the line between AI and non-AI assisted art? What about a gaussian kernel filter? How about adding a saturation enhancer? How about putting a few of those together and let's play with the parameters a bit for fun?
> Lexica, a search engine that solely indexed images from Stable Diffusion’s beta tests in Discord, has over 10 million images in it. It would take a lifetime to explore everything in it, a corpus made by a relatively small group of beta testers in a few weeks.
There are many interesting points, but this is a key one I think. The sheer volume is astounding, like the media flooding of the last decade taken up exponentially.
Perhaps possible to enforce now, but they're fighting a losing battle. AI art will ultimately be indistinguishable from human art except by being consistently better; they're relying on the good faith of contributors. And anything that relies on the good faith of 100% of humans is something doomed to failure.
These tools are only weeks or months old. At the moment they’re a novelty.
‘A picture of a dog riding a skateboard by James Gurney | Trending on Artstation’ is clever and astonishing, yes, but we’ve nowhere near scraped the surface of what they can do or how they can be used as part of the creative process.
It’s going to take time and a good few rounds of development and pushback, but eventually we’ll wonder how anyone did anything without it.
Current trends in AI art are pretty boring, but once we properly get the hang of it it’s going to be fascinating.
Reminds me a lot of the early days of Photoshop - adding a filter was cool and novel so most people added a filter and called it a day. But over time people have figured out how to do all kinds of creative things with it.
As an artist, AI tools are like having a paintbrush that’s possessed by a drunk genie and that’s very exciting.
This 1000%. I'm currently working on my own slightly more complex/customizable interface for Stable Diffusion because I firmly believe that the value lies not in simple prepackaged text-to-image, but in the wealth of information, structure, and hierarchy that is compressed into its 4GB latent space. SD encodes only 4GB of information, but that information is so highly structured and compressed that it represents a sizable chunk of all images on the internet. The true value Stable Diffusion brings to the table is the sheer level of semantic meaning encoded into the vector spaces it samples from to generate images. The fact that it's open-source means anyone can write meta-algorithms to walk through and explore that latent space, too, and as Stability and other companies start releasing better and better models, the latent spaces of those models will become less and less entangled and more and more meaningful and——
We're on an exponential curve and it's so exciting.
Using any Photoshop tool still requires some work. AI is an order of magnitude more multifaceted and autonomous. What happens when we realize that AI art generators consistently and instantly bring forth better contributions than what we can make ourselves over a much larger period of time? The artist will look at their paintbrush and wonder what the point of holding it is when the drunk genie is doing much better on their own. There will still be artists, but there will be space only for the most exceptional of them.
Art software is similar to cruise control. AI art will in a few years become the equivalent of a fully perfected autonomous car that is safer than a human driver in all but the most contrived scenarios.
My phone camera can take much better pictures of things than I can draw. Doesn’t mean that my drawings have been invalidated.
Human art is going to adapt, similarly to how it adapted to photography. If a camera can make perfect reproductions of reality, AI is a camera of the imagination.
I wonder if there’s going to be a trend towards craft -tangible, clearly handmade things that an AI can’t do.
Photography is a distinct medium with its own expectations. People looking for 2d non-photography art generally aren't looking for a perfect representation of reality, so photography does not supersede what they want to see. In fact photo-realistic drawings are viewed as a bit of a meme, especially if it's a picture of some pop-culture character like the Joker.
It's worth asking at this point what exactly the adaptation will look like. I definitely agree that there will be more interest in non-2d, non virtual art. But for 2d/virtual artists? We need more than just an "underpants gnome" sequence where adaptation is considered a given.
The work of those artists won't be invalidated in a philosophical sense, but the competitiveness of the space will skyrocket. The capacity to live as an artist will be much more restricted and it's already a struggle to begin with.
> Left unchecked, it’s not hard to imagine AI art crowding out illustrations that took days or weeks for someone to make.
This reminds me of what someone from the 1800's would say:
"If left unchecked, it is not hard to imagine photography crowding out paintings that took days or weeks for someone to make."
Or several thousand years ago when the pottery wheel was invented.
Or when humans learned to cast metals instead of having to chisel stone.
Art is the expression of human creativity. Art transcends the medium. In addition, making art more accessible to more people is a good thing.
There are probably a lot of people who did not consider themselves artists, but with creative use of prompts and img-to-img can actually create stunning works. That is a good thing, not a bad thing.
> "If left unchecked, it is not hard to imagine photography crowding out paintings that took days or weeks for someone to make."
The difference I see is that photographers don't try to pass off their works as bona fide paintings without telling anyone. Even if they did, it's usually easier to discern whether something is a painting or a photograph than with AI, because there's usually no need to aim for replicating the exact look of a photograph in a painting.
AI artists submit their work as digital art, and this distinction is far harder to make owing to the sheer sophistication of the former. The problem is that both use the medium of digital art, but the provenance is different, so there is still a desire to classify them differently. And as discussed, the huge output of these programs makes the need more urgent.
The interesting questions in this debate to me involve asking what happens when you take the humanity out of art creation, and how the underlying humans will react to that. Even photographers have to use their legs to go outside, train their eyes to find something interesting and press the shutter button with a fleshy finger, all while having the final output able to be visualized in the process. I'm not sure if "art" has ever been decoupled from humanity at this large of a scale in human history before, and that could signal a significant cultural shift if so.
The cultural implications of someone or something being associated with "AI" will outstrip any advancements in technology. Many artists are revered because of the culture surrounding them, not just the specific combination of paints they slathered on a canvas. It is that culture that I think will undergo turbulent changes as a result of the clashes between technologists and the larger art world.
If the culture decides that AI-generated art is plagiarism, then this debate will never end.
Also, asking "what prompt did you use" is probably going to become an insult with the implied intent of dehumanizing its target in the near future, when applied to certain art spheres.
It depends what the point is. You wouldn't let a motorcycle participate in a horse race. I think there's a place for AI art and it's fine if "manual art" remains the same. The difference can be arbitrary at times, but art often is.
> There are probably a lot of people who did not consider themselves artists, but with creative use of prompts and img-to-img can actually create stunning works
That doesn't make anyone an artist just like filling in a coloring book doesn't. You're following rigid guidelines and given a very small amount of creative choice. I get it, it's a very powerful feeling watching a computer approximate a prompt to great effect but it's not artistry.
When digital effects and CGI began to replace stop-motion monster effects in movies, the stop motion artists screamed that art made on a computer was not art, and that their industry would be lost.
Nowadays we know better, not only is digital art accepted everywhere on earth, it's almost expected, and as for the stop-motion folks, they're still in business! Movies still come out every year, some of them with huge budgets (like Coraline) that use stop-motion because it has a certain look and feel that is all its own. AI art will only get better, but so will the ways we use it, and the people who are actually ARTISTS will be the furthest ahead, and leading the way.
It's going to be interesting to see how this is going to get enforced in the likely future in which AI art becomes in every way indistinguishable from human art.
They can just focus on obvious cases: people that upload new images at an astounding rate, people who forget to remove the AI watermarks, and people who simply admit to using AI. No need to turn it into a witch hunt where you accuse everyone of using AI.
It will 100% turn into a constant witch hunt - I can absolutely promise it.
Want an easy reference? See First Person Shooter games.
Turns out that once the computer can mostly do the job, you will see:
1. Constant use of the computer for that job
2. Constant accusations of using the computer in places that restrict it.
See basically every online FPS community that exists. And that's with incredibly invasive software running on player machines explicitly to stop them from using the computer to play - and People both still do it (in HUGE numbers), and the communities constantly bitch about the "rampant cheating".
One way to take down someone you don't like will be to accuse them of using AI. It's an accusation that's impossible to defend against; you could require people to provide a video showing them creating the art, but that adds enough friction to be the death of a community and down the line will itself be something that could be generated.
They're just tilting at windmills. It's already difficult enough to catch cheaters in chess, where there are, at most, a few dozen candidate moves in any position. This is a space with infinite candidate moves.
Best case scenario is that they catch a few people who have no idea how to create art otherwise. But, short of an exam-like proctoring system where art can only be created under complete supervision, I doubt there's anyone in the whole world who can reliably distinguish a creation that is 90% AI / 10% human from 100% human.
I agree it will be hard or impossible to catch cheaters. But you get it backwards: AI generated art probably is more easily detectable. A chess game can be represented in a relatively small number of bytes, while artwork must represented with a much larger number. That means more opportunities to find an unintentional signal of provenance (e.g. artefacts that aren't readily noticeable to humans).
until ai actually starts procedurally creating images with bespoke tools rather than blending pixels it will in fact be fairly easy to distinguish a made image from a generated one for the same reasons you can usually tell if someone painted over 3d or used a photo base. ai has a lot of dalle-isms and artifacts that i cannot see going away any time soon.
on top of that you could simply require posted images to come with a process timelapse. heavypaint and a few other pieces of software already store those in the png data.
It seems like the actual problem they're trying to solve is volume, so why not just rate limit submissions? They just need to look at their median submission rate prior to a few weeks ago and then make that the submission rate limit.
The problem being solved by these decisions is that AI art is a visible threat to the livelihoods of the people the site is designed to support and uplift. Art sites have already had rate limits for years.
These communities are explicitly saying "no" to this disruptive technology. If you want to go make a new gallery site for showing off your prompt engineering that's your own lookout, have fun trying to build a community and figure out how to keep the site funded. But you can't come spam our existing communities of people who've dedicated decades to their craft with the messy, incoherent results of the whole six months you've spent learning to be a "prompt engineer".
> The problem being solved by these decisions is that AI art is a visible threat to the livelihoods of the people the site is designed to support and uplift.
That's not a solvable problem. Art will be made by computers, it's inevitable. Eventually it will get good enough to be indistinguishable from human made art. It's already good enough that a human can use the AI to get 90% of the way there and then finish it themselves.
> Art sites have already had rate limits for years.
Clearly not good enough if every one of them is saying some variation on "our site is getting flooded with AI art"
> These communities are explicitly saying "no" to this disruptive technology. If you want to go make a new gallery site for showing off your prompt engineering that's your own lookout, have fun trying to build a community and figure out how to keep the site funded. But you can't come spam our existing communities of people who've dedicated decades to their craft with the messy, incoherent results of the whole six months you've spent learning to be a "prompt engineer".
I agree, people shouldn't be spamming crap directly from the AI. But good luck keeping out the people who tweak the AI generated art to make it look pretty good. A smart site would embrace this kind of art, perhaps make a new category for it. But trying to ban it from the community is a lost cause.
If every art site bans AI spam and there's nowhere left to post it, and a healthy market for young artists to do beginner work for beginner prices, then the problem sounds solved to me.
And none of this is inevitable. If everyone trying to improve this technology hears from artists who are intensely unhappy about it and quits working on it, if AI art becomes inherently tacky so that it's embarrassing to work on and funding dries up, then progress on it stops. We were all gonna be using full color e-paper displays by now but all the funding for that dried up because emissive displays got real cheap and nobody wants to bother with reflective display technology any more.
I just can't see every site banning AI art. As long as there's a demand for a place to post AI art there will be places where that happens. I can't say how much it will impact beginner prices, but I'm 100% sure that the world isn't going to stop developing or investing in this technology because it makes artists upset.
You're right about one thing though, economics are a massive driver in results. If human art became cheaper than AI art I'd put my money on human art, but I don't think human artists want to work for less than AI does.
I'm not saying this means artists will never work again. Clearly many will, but I'll bet more than ever it will be only the most talented and creative who can make any money doing it and that it's very likely that most of the artists who make money today primarily due to websites like DeviantArt and furaffinity are going to lose most of that income.
I don't think artists will ever stop doing what they do even if their work ever did become commercially worthless. I don't think artists can stop themselves from doing it, even it just continues to give training data for AI that will out compete them. Their will be more art than ever for humans, but less of it than ever will be made by humans.
I am pretty pessimistic about all this to be honest. My main hope is that people will come to see all this as kind of tacky, and/or that there will be some reconsideration of what "fair use" means that renders existing work in this realm illegal.
Artists need to eat too, if our audience stops paying for our work because this shit is good enough for their needs then we are not going to afford to spend all day making art any more. Unless someone can wave a wand and provide basic income to everyone and I sure do not see that happening in the capitalist hell of the US economy any more.
I think the communities involved here like digital artists, furries etc got a lot of attention hating on NFTs last year, so they are doubling down with the anti-art-generator movement. But I don't think this one will be so successful. Like sure, maybe art community sites could require you to tag prompt-generated art just as a way to label things, but in terms of widespread usage the cat is out of the bag.
Whenever I get on Twitter and see someone incoherently angry about people using Stable Diffusion I'm reminded of that quote from an Android in Blade Runner 2049: “You tiny thing. In the face of the fabulous new, your only thought is to kill it? For fear of great change? You can’t hold the tide with a broom.”
Sure, AI generated art is probably out of the bag, but to lump it in with NFTs and such is an interesting rhetorical move, since the only thing that really links them is their new-ness. That Blade Runner quote can be used to shut down any critique of an emerging technology precisely because it fetishizes new-ness for its own sake, and assigns it an agency of its own, inevitable like the tide. It becomes a cause that generates effects in our lives, and the only question is how we react. This robs us of our own agency as social/political/historical actors who can kill/change/improve/use/abuse technologies and decide how to incorporate them in our lives.
People invoked the same language about things like television, and I find Raymond Williams's critique helpful:
> If the effect of the medium is the same, whoever controls or uses it, and whatever apparent content he may try to insert, then we can forget ordinary political and cultural argument and let the technology run itself. It is hardly surprising that this conclusion has been welcomed by the 'media-men' of the existing institutions. It gives the gloss of avant-garde theory to the crudest versions of their existing interests and practices, and assigns all their critics to pre-electronic irrelevance (1974, 128)
Well in this case the link between NFTs and AI art wasn't made by me, I'm just observing the overlap between the same online influencers/groups doing the same type of activism against two different phenomena.
Sure I think there's always a conversation to be had about how to incorporate tech. Like Reddit banning celeb deepfakes is probably the choice I would have made as well if I set policy on the site.
But what I'm noticing is over the top anger at usage per se. (Someone linked the 'reverse monet' tweet elsewhere in this discussion and the quote tweets there are unhinged.) That's what I'm saying can not be 'killed'--these breakthroughs in large language models
One of the biggest furry art archives has always banned art without much, if any, actual input from the human who created the artpiece. For example, game screenshots are banned almost universally, picture generated art is as well (something like picrew) 3d models aren't, assuming there is posing and such and so on.
It seems a perfectly fine continuation to the existing rules to ban ai generated art.
Also, Art AI can't click on ads, which is probably, ultimately their main concern. If you replace all the monetized users with AI that can't buy things, the economic incentive to keep the website running disappears.
Yesterday I saw a guy saying that Outpaint is nothing new, that photoshop was able to do it for years, I pointed it out that what photoshop does is extremely limited, and the results only look okay in small scale, the difference is night and day.
I think the AI generated art, even at this early stage, is pretty good and have no problem with people submitting it to contests and/or selling it.
Also, as a sometime painter, I would view AI generated art as a great and fun source of inspiration. I'd pick some generated pieces that I like, combine parts from this and that and make it my own.
This stuff may open more doors than it closes, IMHO.
> AI-generated art was not allowed because it “lacked artistic merit.”
As a consumer, what I want is the art, not the merit. Over the last few weeks, I've been able to get images that I simply would have never gotten without Dall-e. Maybe it doesn't provide "merit", but it certainly provides a ton of value that I wouldn't have gotten otherwise.
It should simply be a new "hey guys, look at this cool prompt I came up with" category. Using AI for easy upvotes, unwarranted clout or spam should be banned.
I think the divide is that artists have worked at developing a skill to express themselves (art) while AI artists think they have achieved the same thing by turning a prompt into an image. The former is an act of expression, the latter is choosing something you like out of a magazine. Very different and one is art and the other is not. If you really want to say AI is art, maybe, but the person prompting it is definitely not the artist...maybe the curator.
A hyper opinionated curator that is commissioning work based on very specific criteria, maybe.
I don't think that artist is the right word, but gallery owner or curator aren't quite right either. People are pumping out mountainloads of hot garbage with these things, but others are taking time to think about things like image composition, color and light rendering, etc. that are considered artistic skills. And that's not getting into the people that were already artists and have just added this to part of their workflow.
I do think overall the needle is closer to the curator side of the spectrum, but it's not all the way there, and I don't think it's fair to say that there is no expression in the process, at least for some.
And neither would constitute as plagiarism by the legal definition of it. Yes, it's a problem that they are trained on material without consent but what they produce is something totally new and not even close to plagiarism.
It is? I may be cursed with a good memory but absolutely no "AI art" that my friends passed on to me looked original. In some cases it was painfully obvious what it's based on, in other cases it just felt like i've seen it before.
Legal it may be (although there will be conflicts about machine generated derivative works) but it's also boring.
People are just enthusiastic because it's "do it with a computer". If a young human artist came up with this stuff you'd tell them they're good technically but now they need to develop their own personality.
> In some cases it was painfully obvious what it's based on
Do you have an example of this?
Ideally with a link (for DALL-E 2/Midjourney) or prompt and seed (for Stable Diffusion) to show that it was generated from an incidental prompt - as opposed to uploading the existing image and just using the AI for inpainting/fine-tuning.
I don't keep the image spam from the group chats I'm on, sorry. Man or machine generated. There lies madness.
But I definitely remember being shown something that was the submarine from the beatles album cover with different colours and ... something else that i forgot what it was ... pasted in. The text prompt included submarine but not beatles.
So is most real art. In fact - artists have even created a dedicated terms to this kind of plagiarism: "style study" or even "reference". You can look for this term on YouTube and find a nearly endless result of artists basing art off of the style of other artists. [0]
It is 100% legal for me to blatantly copy the artistic style of Ilya Kuvshinov. I could even sell such work as my own - because it would be. I'd be seen as a derivative or knock-off artist but it can be done. As far as I can tell the real issue is that the AI does it far better - and much faster - than people can. Creating heavily derivative works that, if nonfamiliar with Kuvshinov's work, could possibly be mistaken as being theirs.
That the AI has learned that artist signatures/watermarks are a "feature" of some artwork and so sometimes adds its own has only reinforced people's mistaken belief that it is stealing art when that simply isn't how neural networks like this work at all.
Speaking of - how many people doing style studies of Ilya's work reached out for permission first? I'd wager zero. Because that isn't a thing artists tend to reach out and ask for permission to do. So why should an AI be expected to ask permission first?
Isn't that how most art works in real life? Not every artist will create some completely new style. It's mostly combining bits of styles that you like.
AI art is striking at first glance, but devoid of meaning once you really try to engage with it. It's the meaning people want, but an image's visual impact has been a reasonably cheap heuristic for finding it. The viability of this heuristic is now dead or rapidly dying, and AI generated art is degenerating into yet another form of spam. Relatedly, feeds full of visually-striking images, once impressive, are taking on a "spammy" tinge. New, harder-to-fake signals will be found, and social media will reorganize around whatever those end up being.
I'm personally very happy about it. To me, art is just pleasant, it's a matter of degree. Meaning is nice and all, but not necessary. And with this new way of generating art, people can have beautiful art, cheaply and as much as they want. No more cheap Target art everywhere. It's the replicator of art, nice, beautiful art is now post scarcity and I love it.
How does manual labour add meaning to a mentally creative process? It's not like the AI spits out random images, you have to very specifically tell it what you want in there. What's the difference between "telling" your hands vs. telling the AI?
And even with more traditional art, a lot of people trace something and claim they made it themselves.
However, I do think it's art. Coming up with an idea and then a prompt to express that idea, and then curating the results... That's artistry. It's not traditional, of course.
And with img2img and in-painting, there's a lot more to creating a really good work than just a prompt and curation.
I have no problem calling this art. I'm still struggling with the ethics, though. I'm not sure how I really feel about it yet.
kind of an oxymoron. Why bother having an "art community" when one can just generate all the art at random themselves? That's the whole point of DALL-E and co, an endless stream of generated images... no need for a community for that.
It still takes some skill, patience and luck to get a pleasing result. If you try out DALL-E literally 99% of the images you get from naïvely trying some promts will be complete shite.
How do you deal with the chatbot problem in visual media. It's still Eliza, but without the obviousness of text, how can you tell its just Eliza?
I wonder if their will be a similar evolution in art circles as in chat. What questions and answers about the composition will out the AI generated pieces as mechanical in nature, or if our ability to parse images is different enough from our ability to comprehend speech and text, that it will legitimately be impossible to suss out.
The big problem right now is just sheer volume. Just piles of auto-generated crap burying months of work. That definitely needs to stop, rate limiting or banning seems an appropriate way to deal with the misbehavior.
Hmm I've seen someone draw/paint on both paper and an iPad. It didn't look to me that she was much faster on digital. But then it was original art not copies of something.
To be clear: I fully think that digital art is art.
But it is a lot less effort in a lot of ways. If I'm painting in Photoshop or Krita, I have layers. I can adjust blending modes. I can move them up and down. I can easily ctrl+z a mistake. I can rotate my workspace quickly and easily in ways that would be difficult with physical media. I can easily find the color I want without having to purchase more paints and blend them, or colored pencils, or whatever. I can create brushes and variations on them to match my style. I can go on with dozens of other things.
I happen to think all of that is a good thing, and don't know that comparing digital art to AI generated art is a good analogy, but digital art 100% makes many things you do in art easier.
Photographs are a great analogy. If you work to learn about how to take a good photo, and learn all the techniques around framing a shot and how different shutter speeds affect things you can take beautiful pictures. However, if you tried to submit that picture to someone as a painting, they will cry foul. Also most people don't consider your candids on vacation to be in the same "category" as a carefully crafted (or not!!!) photograph of a landscape.
In much the same way, sure you could dedicate your life to learning the craft of prompting an ML model to produce unique, novel, and interesting art, but these communities are saying "we want these to be considered a different class of art" and as mostly self organizing communities they are well within their rights to do so.
If you are a "digital art" community, clearly you have already accepted some digital tools.
I do think it makes logical sense to try and ban "AI" art if it is not what the community is interested in. It's likely not just about pretty images. They might value the mental process, the individual's style, progression, techniques, and yes, the effort. Someone coming in with generated images expecting to be treated the same is to me a bit like commissioning art from someone else, and then presenting that as your own. The images might be great, but it's not in the spirit of the community. There must be tons of "AI" focused communities for that.
I suppose they will eventually fail at resisting this, but that's another matter.
I don't think that's really fair... I can see the argument for removing AI generated stuff in part because it doesn't require technical artistic skill (whereas digital art still does, in your ability to observe, draw with perspective, etc.)
But that can easily lead to a slippery slope since it requires more technical skill to make physical art than digital (for one, you can't "undo" anything in physical world.) Where do we draw the line?
Secondly, it will just become more and more subtle as art softwares will subtly repackage such AI tools into their art software packaging. I am willing to bet anything Adobe is already employed an army already to figure out how to do so. How would you tackle that?
So if I skillfully combine multiple AI generated images it's stops being art? What's the difference between someone only using that vs. someone taking parts form random images found in Google image search? The latter very common among digital artists.
I don't oppose AI-generated images in the slightest. I love them and I genuinely think they will be a very important part of our future. Effectively, it will take more effort to make a living as an artist, especially a mediocre one, because AI is now a decent mediocre artist. It's a stupendous artist for some prompts, too.
However, I can understand that if I can just spit out image after image, there must be a way to "limit" my submission of these images. In that sense, I "get" banning AI-generated images, as a way to stop spam.
I'm in some Discord chats for various things, and they often have a channel where people can post their artwork. Often, people spend weeks or months on a piece of fan art that's related to the Discord topic, or they draw a picture of the "leaders" for fun.
Lately it's been just getting drowned out by people posting ridiculous AI art where they try to get absurd results. It was interesting and fun at first, but after seeing probably as low as 20 of these images I can tell they're all a similar sort of thing.
I feel bad for people posting their art getting drowned out by this, and these aren't even remotely art related groups.
I've used AI to create stuff like music album covers and posters. It's cool and fun to mess around with, but I think it still takes a real person with an aesthetically trained eye to be fully effective. It does basically what I would do anyway, which would be to composite different images in Photoshop and then apply effects, etc. But you still need to edit for composition and text, and most of the time the images have to be fine tuned, or not at all what I wanted. I can easily see AI taking over most image compositing.
I don't understand the difference between telling photoshop to sharpen an image or remove red-eye and using AI to generate based on a prompt. Both of them are software modifying your canvas.
I'm fascinated by the idea that AI-generated images will eventually be the primary data source for training the algorithms themselves. Kind of like a giant positive feedback loop. Eventually the "art" might not be about generating a single image from a prompt phrase, but influencing art through the generation of the most successful algorithms.
I recall some time back the "photo of the year" with some esteemed organization looked like something you'd take at random. Just a few people doing various things. A snapshot, the sort you'd take back during the film era when you'd load film then click the shutter in case the first photo had a light leak.
Typing a prompt into DALLE does not take effort and it does not make you an artist.
Don't kid yourself that the people generating art and submitting it to these sites are the same ones writing and training the AI.
"Pressing a shutter button on your camera and tweaking a few parameters in settings does not take effort and it does not make one an artist."
I think we have established by now that photography is art, so your argument doesn't feel like it would stand the test of time much. Just like the photography one didn't.
I think you're just proving my point actually, being able to take photos doesn't make everyone an artist or every photo art. To take truly great photos you still need to understand basics like composition and lighting (just like illustration and painting) which do take effort, and your still not guaranteed to get good photos.
A camera, like AI, is simply a tool and it's no doubt already helping artists, but to think typing a prompt into a website somehow makes you an artist on its own is foolish.
In that sense, yeah, fully agreed. Someone typing up whatever in the prompt is about as much of art as someone taking a random selfie in poorly lit environment on their smartphone.
However, there is definitely nuance to generating art using AI that is similar to photography as art in a way. There are massive blog posts and writeups that exist already, talking about different techniques for crafting prompts and fiddling with the model to achieve different results. Stuff like how ordering of the words matters (e.g., the order changes how much the model weighs a word), how the choice of different samplers affect the outcome, which words tend to be strong modifiers (e.g., if you want something looking like a real photo, add a specific camera model name), the number of iterations and guidance scale parameter choice, etc.
That's barely scratching the surface, given the whole thing became available less than a month ago, and people discover all those methods and techniques in real time. I am willing to bet that in a year, the complexity of a "good prompt" will be lightyears away from where we are now.
It's like someone invented a music instrument/artmaking tool, except it behaves a lot like a blackbox. So instead of developing techniques from the base knowledge of how the tool was created, we have to test and try bajillion different things to figure out how to get what we want.
And no, I am not trying to make a claim that generating art using those AI models is an equivalent to creating an acrylic painting. Just like i wouldnt claim that taking a great (in terms of art) photograph is an equivalent to creating an acrylic painting. Both of those stand in classes of their own.
The photographer definitely has more agency over what they are shooting than a programmer does over what an algorithm produces. When I take a photo I have a image in my head I am trying to capture; when I train an AI to do the same I don't have that image, its just generating a bunch of images based off an idea and I select the one that is most appealing. I'm sure there are people doing tuning that is much more involved, but at a basic level it's not the same thing. Doing photography or drawing are much more emotionally rewarding than generating images with code, at least for me.
They aren't banning AI art because because it doesn't take effort. They're banning it because once that work has been done, you can create a huge number of images which creates problems for the community.
Yes, creating the actual ML model and curating it's input takes effort, a lot of it in fact. Taking someone else's ML model and shitting prompts into it does not
You can say that AI art is bad. But you can also say that normal people have bad taste. So the two meet in the middle and the normal art-consumer is satisfied with his junk.
It's like fast food and normal people. Everybody knows it's crap but they eat the stuff by the megaton.
The argument againt AI art is an argument against popular markets, basically.
This will become even more interesting as digital art softwares start including "AI brushes" and such that include subtle tweaks to the established process. AI generated art made in 40 seconds could be banned, but what would happen to the AI assisted works, churned out by an artist in 10 minutes instead of 10 hours?
I'm waiting for the first case of a human artist being banned because their style was "too similar" to AI-generated art. I think that'll be the beginning of the end for the communities that do these kinds of bans.
That's shortsighted. I'm sure they'll come around. There's already been some great demonstrations of artists using ai to enhance their visions instead of being just a simple output.
How will they know what is and isn't AI-generated? And when they cause a false-positive how are the "real" artists able to verify that their art is legit?
My normal artwork (digital) is 9000x9000 pixels up to about 14400x14400. Good luck getting any of them to make something bigger than 1024x1024. Sure, you can use Gigapixel to upres things, but in the end it's basically impossible to make a decent artwork using ML at the moment. I only use Dall-E for offbeat ideas for geometric abstractions, so far Stable Diffusion sucks for that purpose.
Of course when you upload at no one wants you to do the full resolution image; at 1024x1024 you might be able to fool people with something ML made.
I believe most of these models only use 1k images so they really can't make anything bigger yet. Maybe someday but not today.
Have you seen the 'in-painting' technique yet? You can write prompts for each element of an image you want, erase the parts you don't like, position them on the canvas, and make SD fill in the blanks.
Problem is the computing power necessary to implement those bigger models increases by some large value as well. An 81MP images is 81 times bigger than a 1MP image. Not sure exactly how the generation is implemented, but I am sure it probably becomes unaffordable if its linear.
It might require dedicated hardware. That only really becomes possible when you've proven the idea, but ASICs for cryptomining, TensorFlow, etc are quite real. There's no reason why dedicated hardware for training Stable Diffusion couldn't happen.
It's mostly a VRAM limitation at this point though, since even a 81x slowdown with something like Stable Diffusion would be perfectly acceptable to produce the final high-quality render.
Which is to say, it sounds like the solution is dedicated GPUs that focus on VRAM over speed.
Is it possible to automate detection of automation? If you can create an algorithm that detects the difference between AI art and human art, wouldn't that algorithm itself be able to create a new AI art algorithm that it itself could not detect?
Like this isn't gonna cause a backlash against the half assed "Artists" that modern art is. If AI can do everything better than humans, these luddites either can evolve or face the same fate the horses and loomers did when the industrial revolution came.
I'm not surprised more and more art communities are banning AI images, 99% of them are dull and ugly. Definitely not something I would print out and hang on my wall. I tried playing around with these models, but I was too bored to bother futzing around with prompts to try and make something I liked.
But I have seen some cool things made using these models by creators who put more effort in than just writing a prompt to get back an image. I can definitely see how these tools can enhance human creativity but like most art the results are just bad.
At least with some human art even if it's not aesthetically pleasing you can appreciate the thinking or time and effort put into producing a piece of work. When you go to an art museum you can learn about history, politics, culture etc even if you don't appreciate how something might look.
Duchamp poked the art establishment in the eye with his 'readymades' (urinals on the gallery wall, etc.) Warhol with his 'factory' making silkscreens.
The 'reverse Monet' (too lazy to link) that caused such an uproar when it was shared last week was ironic – apparently people aren't aware that the impressionists themselves were considered 'not artists' by a lot of people at the time.
The camera and photography made many painters sit up and say, 'hey, it can't just be about "image" anymore' – enter abstract expressionism, fauvism, cubism, etc. Work that shifted the focus from the image represented in the painting ('a window into a world') onto the surface of the canvas itself: Rothko, Albers, etc.
The thing is, people have wild misconceptions about what this technology does. Many clearly think it's a photocopier. As if StableDiffusion is hiding copies of all these works inside. When in fact what's been done is to take all of this online knowledge and create an incredible new tool with it. Like a music synthesizer times a million.
It's understandable why this technology makes people uncomfortable. There are legit fears about displacement, economic uncertainty, etc. But we are fundamentally tool makers – it's arguably intrinsically part of what makes us human (borrowing from Carl Sagan's "Dragons of Eden").
If I were back in the art department last century, I'm pretty sure we'd be really excitedly discussing how these new tools will enable entirely new forms of expression. And sure, there'd be the old guard saying it was awful, and 'not Art'.
This is a really interesting debate – I just wish it scratched deeper than the surface more often than it does, with a bit more historical context.