In that dialog box there's a text field with "modifiers" one if which is called "studio ghibli". I guess that makes for some interesting copyright considerations.
It's funny because art was the activity humans were supposed to be engaged in after AI took over the work we didn't want to do, but it looks like it will be one of the first things to fall to the machines.
But graphics, design, and special effects were some of the first areas heavily influenced by digital technology, so it’s not super surprising that AI is showing early signs of future impact in the visual arts.
It’s becoming pretty clear that some people are much better at harnessing these emerging tools than others. Some people are already billing themselves as “AI artists,” and I predict that many will find success in this new art form.
Parent is principal AI scientist (Id say a $1M+ per year position), his net worth (in billions) is very likely function of the success of these tech.
It’s becoming pretty clear that some people are much better at harnessing these emerging tools than others. Some people are already billing themselves as “AI artists,”
This is evil bullshit. Whatever creativity or value rather an artist creates with these "tools" will be sucked out by the AI directly from the photoshop canvas and recycled the next day except the artist wont be necessary this time.
and I predict that many will find success in this new art form.
Until the pupil has surpassed its many masters obviously.
The artist is necessary for creating anything legitimately new that doesn’t suck. The critic is necessary to filter out all the crap the AI would be happy to create. Look at all these examples where the text prompts literally say which artist’s visual style to copy.
Making derivative artwork is in the process of becoming completely commoditized. That’s good in the sense that it allows novices to create visual content that they previously could only imagine (presuming they can adequately describe it).
But this also means the half-life for visual imitation will be drastically shortened, because even viral new styles will be done to death by artful copies in a matter of weeks.
Is XKCD art? Visually unchallenging (?), but certainly creative and thought provoking. What we’re mostly talking about at this point is giving kids (and some adults) who can’t draw well yet a chance to express what’s in their heads onto the page.
It’s just a blob of paint until someone says (with conviction) that it’s “art.”
Creating compelling results which connect with people requires practice, patience, the willingness and ability to explore, a whole lot of cherry picking. Iteration.
There will have to be protections for content creators, which will have to evolve to accommodate AI’s increasing capabilities to mimic, some might say “steal,” the style and character of an artist’s work. It’s simpler in the case where the AI clearly imports entire assets from a single source, much harder when it almagamates tiny fragments of a large number of different artists’ works. Some kind of tax may be necessary to ensure that human creators of source materials are fairly compensated. Nothing happens without their work as the basis of creation.
It’s just a blob of paint until someone says (with conviction) that it’s “art.”
You perfectly know fine art isnt the issue here because its a small industry anyway and you guys arent investing dozens of billions to take over a niche industry.
For those who dont know what we are really talking about here: it's the illustration and entertainment industry, which combined are probably well over a trillion dollar industry.
Creating compelling results which connect with people requires practice, patience, the willingness and ability to explore, a whole lot of cherry picking. Iteration.
It doesnt. And you know it. It only takes taste. Which is very easy for an AI to acquire with intelligent supervision. It will be way better than the 99 percentile of artists very quickly. The rest is monkey work and is worth nothing.
There will have to be protections for content creators, which will have to evolve to accommodate AI’s increasing capabilities to mimic, some might say “steal,” the style and character of an artist’s work. It’s simpler in the case where the AI clearly imports entire assets from a single source, much harder when it almagamates tiny fragments of a large number of different artists’ works. Some kind of tax may be necessary to ensure that human creators of source materials are fairly compensated. Nothing happens without their work as the basis of creation.
One of the first things I noticed on the SD website is the cleverness of the bullshit. With "ethic" and stuff everywhere. Did you wrote it? Are you CEO?
There wont be taxes mate, there will be questions :)
“You perfectly know fine art isnt the issue here because its a small industry anyway and you guys arent investing dozens of billions to take over a niche industry.”
Seriously, I don’t know that. According to Statistica, the total value of fine art transactions was over $65 billion in 2021. But fine, let’s take it as stipulated: commercial art is a far bigger market.
If there really is that much at stake, change is inevitable.
“It doesnt. And you know it. It only takes taste. Which is very easy for an AI to acquire with intelligent supervision. It will be way better than the 99 percentile of artists very quickly. The rest is monkey work and is worth nothing.”
You are (aggressively) psychic! As an AI practitioner, if I could so easily bottle up taste and sell it, I surely would. Supervised learning requires a source, and current methods require vast (and thus expensive) corpora of training data. The better it understands what is “good” by the standards of that data set, the less ability it has to generate novel work and redefine what is, in fact, good.
Visual style, in particular, is subject to fashions whim, what ML practitioners call “concept drift.” What is exalted one day, is scorned and ridiculed in another.
If taste were so easily captured in a model, Hollywood would never have to worry about making another bomb. Game studio executives could sleep blissfully. A vast amount of money is lost due to bad design choices, across countless industries. Frankly, I think a Taste Oracle is worth more than an AI Snuffaluffagus.
“One of the first things I noticed on the SD website is the cleverness of the bullshit. With "ethic" and stuff everywhere. Did you wrote it? Are you CEO?”
I am just not that clever. I see all that as CYA. The fact that they released everything, code and models, undermines the notion that they expect to corner a multi-trillion dollar market.
Paranoia, anger, aggression, maybe this is appropriate in the face of unknowable change. But people are divided at the outset of this revolution, and remain conquered by the future Mars dwellers.
To clarify, I work (Earthbound) in Computer Vision and Perception; my preference is for concrete problems.
‘Fair point. I am not judging you Im just trying to educate an uneducated reader and call a cat a cat: there is no "AI artist" thats a lie. Only "AI trainers"’
If people can make art out of literal garbage, I think they can make art by manipulating a Turing-complete class of machines (recall that LLMs can perform program synthesis).
I agree that naive and even not-so-naive usage of AI tools will mostly be artistically superficial, but I believe that (some) people will break through the creative limitations inherent in the copycat nature of this machinery.
Fair point. I am not judging you Im just trying to educate an uneducated reader and call a cat a cat: there is no "AI artist" thats a lie. Only "AI trainers". And youre going after the entertainment industry, not art galleries. And that new pen you give the kids come with a special paper that copy and exploit their creations.
Now maybe thats okay. Maybe there is no need to lie about it. If there is paranoia, it doesnt come from me I believe.
If your client require to illustrate specific situations with specific details with specific camera angle and lighting... Word to image AI generators wont be enough. You need skills and experience for paint specific tasks.
Not to give credence to this space, but NFT art shows presenting Pixel Chill Monkey #9491 made with a coked-out doll dress-up game is a seemingly much lower bar
These image-from-prompt generators are themselves just a particular type of generative art.
I think it's obvious that the output of these AIs can be considered art. There is human input at multiple levels - the code, the artists whose works it was trained on, the researchers that scraped/pruned the training set, and ultimately the crafting of prompts by users to achieve a desired output.
What is not obvious is: among those human inputs, which are actually responsible for imparting artistic value on the resultant image, and in what proportion.
It's not that simple. The current generation wave of so called "AI tools" such as GPT-, DALL-E and the rest generate fantastic results because* someone did the enormous work of scraping a huge amount of someone else's work (texts, images). Of course you'd argue humans do the same: they absorb a huge number of inputs and generate content based on that. However, there is something else in automated solutions like this one that places tools like Copilot on the verge of more or less sophisticated plagiarism. And when you look closer at the images, you discover oddities just like in the faces of AI-generated people.
Real art, on the other hand, is all about innovation and exploration. That's one of the reasons there are fewer and fewer traditional paintings in modern art museums. The current generation of AI is still very much a copycat, not a creator. I believe this might change, but we'd need a complete paradigm shift for that - ML is almost useless as the core of such solution, as much as generative art algorithms.
I disagree. I believe that art is eternal, but we still fail to grasp what it means. I don't think art is limited to pretty pictures and poetic prose. For example, if a human laboriously produces a work that they could have simply asked an AI to produce, isn't that starkly superfluous? To me, a performance art.
But at least photography created new roles for photographers. AI art promises to replace many creative roles for humans entirely, without creating any new roles.
These AI art tools model distributions based on sample data. They don’t create the underlying distributions; they need to be given (many) samples from the source distributions.
You need people even just to say if something is good or not. Standards will rise considerably as AI tools mature, and peoples’ skills with the new tools grow. Things get old real fast and people get bored quickly.
Here’s a new role: creating trademarked artistic styles. People will become tired of what eventually becomes seen as a limited set of “AI styles,” which are actually distilled from the techniques of a relative handful of human artists. So they will seek out “Natural” artists who are able to create new artistic styles. Also, the Disneys and the mega-studios will need and want to create distinctive and commercially protectable visual styles.
The legal framework and case law will have to be worked out. Think of it as “Look and Feel” on steroids. One thing’s for sure, it will create significant roles for creative IP lawyers.
The examples you gave sound like they would only be for the top tier of today's artists though, or a small number of AI taste supervisors, not the general workforce. A company would need to hire far fewer people if all they need to do is feed a new style into the image generator occasionally. I also don't see that as a new role created by this technology, since people are already doing that - those are just a few people who might not get laid off.
If standards rise considerably it could help, but I'm not really able to picture what that would look like, in a way that would mean more roles for people. Or if the tech gets stuck where it is now and never becomes truly useful, then I could see AI as just being another tool.
The invention of recording devices gave rise to the recording artist. Only a relative handful of people can be super successful at this, but there is a greater population of working musicians and singers.
YouTube and TikTok, smartphones allow millions of people to become influencers. Some people make millions playing video games live online on sites like Twitch. Only a handful become rich through new media, but it seems to allow a fair number of people to make a living. Certainly there are far more minor celebrities than ever before. Even (really talented and/or self-promoting) chefs are celebrities now.
I’m sure people will take classes, hire instructors / tutors, buy books, watch YouTube videos on how to get better and/or more efficient with AI art and design tools. It’s not enough to teach people new technical skills. You also need to teach them how to monetize those skills.
The key takeaway is that the barrier to entry has dropped through the floor. There will be an explosion in visual content. In a world where nearly everyone is at least an amateur content creator, and probably at least internally a content creator at work, there’s going to be a skills pyramid and strong demand for legitimate talent.
Well, I agree that some technologies do justify themselves by creating more jobs/roles than they take (I think you're overstating the number of people making a living as social media streamers/influencers though). But the question is whether AI art specifically is going to do that, or if it will simply remove jobs and not add any back. People might still buy books and train themselves, but if the demand for art can't scale up, they won't be able to find good jobs or creative fulfillment.
I didn’t state a number of people making a living on social media. Some people are clearly making a fortune; how many people who are making a living off it, I don’t know. Enough so that I typically have no clue who some random Instagram, TikTok, or YouTuber with a million followers is.
Historically, new technologies create more jobs than they eliminate. Books used to be copied by hand, by the relatively few people who could read and write. The printing press eliminated a lot of the bread and butter scribe work.
Employers love finding ways to eliminate labor. Autonomous vehicles promise to eliminate millions of driving jobs. But eliminating labor is not an end, it’s a means to an end, which is increasing profitability - a function of productivity.
Employees have stayed ahead largely by boosting their productivity faster than labor saving technologies eliminated jobs. The same technologies which eliminate jobs tend to simultaneously increase the productivity of other jobs. Employment is at an all time high.
I have no idea what all the new jobs created around AI art and design tools will be. I presume it will create a new subfield of law. There will be new laws and regulations, regulators, some compliance burden on businesses (e.g., are they illegally exploiting unlicensed commercial content), auditors. There will probably be reviewers and editors for AI generated content. There will probably be considerable AI infrastructure, consultants, support personnel, administrators, AI model and data providers, annotation services.
For most office workers, the level of graphics and visual elements in reports will probably go way up. PowerPoint will be a distant memory, but expectations regarding presentation quality will seem insane by current standards.
> I didn’t state a number of people making a living on social media.
Sorry, I misread your implication then. Twitch had a leak a while back and something like 4500 people were making over $50k USD, which is not nothing, but also not that practical as a career strategy.
Hopefully you're right that AI will open lots of opportunities for everybody, and hopefully those opportunities will be worthwhile. I know historically technology hasn't brought about the end of all jobs, but AI is a pretty new concept and we can't necessarily scale up forever. I think it's reasonable to be concerned about the number and quality of jobs remaining. It does seem like some of the jobs that people really want to do, are some of the ones AI could be taking.
But I don't really know one way or another, just feeling concerned and a bit bummed out about it. I guess we will see over the next 5-10 years.
Seems like humans are still actively engaged in it. Just place a color blob here and a different color blob here assign those color blobs with different tasks and tell the AI to fill in those blobs with different descriptions.
Agree. For me, as not very skilled artist (but still able to make living by creating art) it opening new layer to creative process. I can spend more time by thinking about content than just drawing or spending hours for researching specific images.
You might be right (at least until the AI improves a bit), but even so it promises to make artists more productive to the point that we will need far fewer of them. Unless we are able to scale up the amount of art we need by a huge amount.
This is a smart move and goes to show people move so quick on this. I wonder how long before these models overtake stock image websites, get used in games or video production, generate T-shirt prints and whatever other usecase people can come up with.
On that note: Iirc back when I learned about GANs, when they were still rather new there was a clear path towards improvements of them by making the model bigger and feed it _much_ more high quality training data. When I look at the outputs of stable diffusion or Dall-E there are still often visible artifacts and most prominently faces are often weird. Is there a clear path towards improvement here aswell or are we hitting a wall with the "just more" paradigm somewhen?
GANs have a key (learned) loss called a discriminator which tells it whether or not the generated output looks real or not. GANs took many years of research to get to the point where they could get to the point where they could generate realistic, full sized images. Progress in diffusion models has been much faster.
But you can add additional guidance to diffusion models, for instance, (possibly pre-trained) discriminative models which detect if certain objects classes such as faces look abnormal or malformed. We’re still quite early in the evolution of this family of models.
When my friends and I play D&D, usually the host will quickly sketch maps on the fly (just, like, a floorplan or something). This tool seems like 1-2 iterations off from being perfect for that sort of thing.
I knew that eventually Photoshop would get some plugins utilizing AI image generation but I never expected it to happen this quick. From a brief glance at the demo (see reddit link from simonw) it seems to be of decent quality as well.
According to the reddit comment by alpacaAI the plugin uses cloud/hosted solution, however [1]. Locally-powered generation to be added as a feature later it seems.
Still, pretty amazing (or terrifying for some) at how quickly we are advancing with this diffusion-based tech. I wonder what is going to be the next big AI method for creative-work generation?
While the video someone linked here looks incredible, the website has information density around 0. It’s a little annoying to having to research something instead of just reading about it on the project website.
I understand you’re in a private beta, but a demo video, some example images, and a little bit of descriptive text would do wonders in improving the website.
Your second link validates my statement. You can absolutely trademark generic terms. There are limitations though.
Try to create a smartphone or computer company called Apple and see what'll happen. You can't trademark the word Apple for your apple farm because other apple farms must be allowed to use that word to describe their product.
That other companies have Apple in their name is not surprising. A trademark is not universal. There are so called classes and when you apply for a trademark you have to specify which classes you want covered.
Source: I went through this process and applied and was granted a trademark for my company. Other companies hold a trademark on the same term but in different classes.
