Me, Myself, and AI
Created with Stable Diffusion
Last week I wrote about using AI-generated art. Mostly, I focused on what it is and where it fits in production… right now. Truth be told, Artificial Intelligence technology is moving at a very fast rate, and I realized that ignoring it would be a mistake. Whether we like it or not, AI is here, and they’re coming for you.
Either we learn to adapt to it, or we get passed by. But what about all the legal and ethical baggage surrounding it?
I have spent a good amount of time lately working with Stable Diffusion, an AI art generator that can make images of just about anything, in any style, and in a few seconds. No, it isn’t perfect. In fact, I have been running into a lot of limitations with it. But I have also been seeing a lot of potential, and learning a lot about what it takes to make decent, or even good, AI art. Let’s not get ahead of ourselves though. Before we start asking why, when, or how we can make AI art, let’s first look at if we should make AI-generated art.
Is AI-Generated Art… Art?
One of the most cliché ways of deciding if something counts as valid is to define it. So let’s do that first and get it out of the way. According to the Merriam-Webster dictionary, “Art” is defined as:
1: skill acquired by experience, study, or observation
2: a branch of learning one of the humanities
3: an occupation requiring knowledge or skill
4a: the conscious use of skill and creative imagination especially in the production of aesthetic objects; also works so produced
4b: (1): FINE ARTS (2): one of the fine arts (3): one of the graphic arts
5: (adj) produced as an artistic effort or for decorative purposes
Taking these into account, one could argue that an AI like Stable Diffusion is not really acquiring a skill, but rather running an algorithm to mimic the skills of others. It’s also unlikely that these AI are putting any “conscious” effort or “creativity” into making these since they are not conscious and are not proactive. They only do what you tell them when you tell them to. The only thing you could argue is that the works produced can be used as art, but again, with intention by a person.
But so what? Art or not, these pictures are still being made, and the AIs are getting better at it. Is it something to fear? Are we all going to lose our jobs to AIs? I don’t know, but I doubt it. I can’t say jobs won’t be lost. Even back when the industrial revolution came about, a lot of people lost their jobs, but they didn’t stay unemployed or unskilled. And working in most industrial fields require a high level of skill and training. It just meant a change in what people were doing. And that is likely to be the case here as well.
The reality is, AI-generated art is another version of CGI, or Computer Generated Imagery. The computers we all use don’t get the credit for making the art we produce, be they illustrations or films. A computer is a tool, a tool designed to help the artist. What I am able to do with my computer today was impossible 5 years ago, and unimaginable 20 years ago. But let’s take a slightly deeper dive into this by looking at the history of art and the controversies surrounding technological artistic development.
The Controversial History of Historic Controversies
The earliest cave paintings were merely sketches, drawing that depicted the iconography of animals and people. They portrayed ideas rather than depicted the reality of the world around them. There were no photorealistic cave paintings of oxen. But the desire to better reflect reality pushed art forward, and eventually, humans were creating highly detailed and realistic carvings and drawings. The ancient Romans excelled at realistic sculptures, something only achievable due to advances in technology at the time, and careful study and observation. But eventually, that wasn’t enough.
Skipping ahead a few thousand years and we run into the development of the Camera Obscura. The origin of this device dates back to the 4th century BCE, when a Chinese philosopher, named Mozi, first observed the phenomenon that allows for the camera obscura to work, the projecting and flipping of an image through a small hole. And there are theories that suggest such technology was used in some paleolithic cave paintings and Neolithic structures. Even Aristotle used one to observe a solar eclipse. But what does this have to do with art? Funny you should ask.
Because a camera obscura is really just a small hole that projects and flips the image of whatever is on the other side, without the use of a lens, it was figured out by the 15th century that you could use it to draw over the image produced to create more accurate representations of things. In other words, they invented a projector and tracing. And it pissed a lot of people off. And it still does.
While modern-day people see the camera obscura as the predecessor to photography, back then, the idea that you were “cheating” to create your art by tracing over an image was very controversial. People still argue about it. Many artists who used optical aids to create their paintings had to hide that fact, and we are still speculating who did and who didn’t. There has been a long intense argument about whether Vermeer used the camera obscura to create his paintings. But, as others before me have pointed out, who cares? The artistic eye, steady hand, and skilled brushwork of the artist are what made those paintings great, not the fact that they used a new technique for the composition. If people thought that tracing an image was cheating, just wait until they hear about the Xerox machine.
By 1806, the Camera Lucida was developed and patented, although similar devices were likely used for 200 years before then. These devices were basically plates of glass that the artist would look through to see a “projection” of an image directly over their canvas, without flipping it upside down. It’s the same technology we use to project the words of a teleprompter right over the camera lens so the news anchor can read the news but the audience doesn’t see a word of it. And, you guessed it, people still get their panties in a twist when they debate who used it to create art.
But all of these techniques provided the means to create some of the most gorgeous paintings. Does it matter that they used something to help them do it? Sure, but not in a bad way. Advances in such techniques are to be celebrated because without them there would be no photography, no film, and no Netflix. That’s right, photography was an outcast in the art world when it was first invented. Developed in the 1820s, it was seen more as a threat to art than as an art form itself. Critics feared it would supersede painting, and thought it lacked “something beyond mere mechanism at the bottom of it.”
And yet today, photography is seen as one of the fine arts. Not every snapshot taken on an iPhone would qualify as fine art, but that is exactly my point. Photography is not the artist, it is the medium. It takes training, skill, and years of dedication to become a good photographer and make fine art, even if the moment of creation is only a fraction of a second. But just taking a picture usually isn’t enough. Artists want to mold the image into what they want, even if that isn’t what they got. The “doctoring” and manipulation of photographs, something we usually call “Photoshopping” nowadays, has been controversial since it was first introduced in… well… the 1800s.
Shortly after the invention of photography, people were already altering photographs, with varying degrees of success. Sometimes it was done for artistic reasons, sometimes it was to add someone who was absent in a picture, and sometimes it was used as propaganda during the American Civil War. These early pioneers in photo manipulation developed many of the techniques people still use today, although in digital form. (There’s a reason the masking tool in software like Photoshop is red, it’s because photographers used red acetate to mask out parts of the images in the darkroom.) It’s likely that, at least in part, the ability to manipulate photographs from the very beginning helped establish photography as a major art form later on. And it is the reason we have blockbuster movies full of superheroes and explosions today.
Cheaters Sometimes Prosper
So when is such a thing ok and when isn’t it? Well… that’s part of the debate. When it comes to art, what is cheating, what isn’t cheating, and does it really matter?
Well, yeah. It matters. I’m sure there will be a lot of differing opinions on this, and probably exceptions to any attempted definition of cheating in art, but there are three things I think are the biggest hallmarks of cheating artists.
First, Honesty. If you are being dishonest about what you do, how you do it, or any other aspect of your art, you are lying, and lying is, in many ways, cheating. You should never be ashamed of the tools you use, the techniques you learned, or the art that you make.
Second, Theft. Not stealing a painting from a gallery, mind you, but stealing the work of another and claiming you made it. I once almost got in trouble at school under suspicions of plagiarism because a friend and fellow collaborator on a project of mine used background art from someone online in an early version of a film I was working on without crediting them or even telling anyone. When the professor saw it, he found the original work and we had a talk. It turned out she was overworked and too busy to make the background in time for the deadline, I had no idea my friend had stolen the art, and she profusely apologized to everyone and had to turn in all her art for the rest of that quarter with layers/iterations to prove she made it herself. Honestly, they were very lenient. Don’t take someone’s art and claim it’s yours. Period.
Third, Fear. Fear is the mind-killer that will make you do stupid and desperate things. When you allow fear of criticism or fear of failure to take over, you’re more likely going to use a safe and established technique in ways you shouldn’t. This isn’t really cheating to make art as much as it is using it as a crutch and stunting your own personal growth. It is cheating in only that it is cheating yourself out of becoming a better artist.
What are the most common ways people “cheat” in art?
Well, if you go old school, anything involving Tracing would be considered cheating. If you can’t draw that kitten without tracing over a picture of it, you can’t draw, or at least that’s the argument. In reality, tracing isn’t cheating at all, except against yourself. If, as I mentioned, you use it as a crutch, you aren’t going to get any better without using it, and all your work will look like it was traced. But if you use it as a tool, and you’re upfront about it, then there’s really no problem.
Here’s a crazy one… using a Reference. Ooooo… The idea here is that if you can’t draw something from memory or your imagination you are a bad artist. I’m going to say right now that if you think this way you’re wrong. No matter what type of art you’re making, using a reference is not only acceptable, it’s expected! You can’t create in a vacuum. If you want to get good at drawing people, don’t just study anatomy, go draw actual people! The ONLY way this is “cheating” is if you are dishonest about it and say you didn’t use a reference. Although really it just makes you a liar.
Using a Grid or Aid is cheating? No, it isn’t. If you haven’t used something like a grid or perspective guide, you’re only cheating yourself out of amazing tools better artists than you use. Just don’t lie about it if you do use one. And don’t be ashamed either.
Some people even think Art School is cheating. The only cheating there is getting cheated out of your money. Not that art schools don’t offer a lot and are worth it in most cases. It’s just that they tend to think artists all make a lot of money, making it very hard to pay back. But school or not, learning techniques is never cheating, that’s just art.
What Does This Have To Do With AI-Generated Art?
Everything. Sort of. Using an AI to generate art is not, in and of itself, cheating. But there are 2 issues most people are concerned about when someone uses AIs like MidJourney or Stable Diffusion.
Created using Stable Diffusion
1: Rights. The pictures they produce had to use “models” that were “trained” on existing art. Most of the time, that art was not used with permission. Basically, it means if someone wants to make a picture of a character that looks like it was made by Studio Ghibli, chances are they didn’t get permission from Hayao Miyazaki to use screenshots of the movies to train the model. The only way this is solved would be to use public-domain imagery or get explicit permission from the artists. Right now many models use a plethora of sources that are impossible to track down which makes using them a great big legal question mark. Who’s art was used to train it? Did they have permission? However, as a counter-argument, any artist that has ever made a drawing in the “style” of Studio Ghibli would be guilty of using these films as references. In this case, it comes down to how we choose to define the use of the original art. Would it be a reference when used to draw a picture, but not when used to train an AI? Ethically, the real problem when you are using the look or style of another artist is trying to profit from it. Plus it begs the question “would you have otherwise paid the original artist instead?” That’s hard to prove, and most people aren’t willing to spend the money needed to get a custom-commissioned work of art from an established artist.
Further, who owns the right to the work produced? The AI? The company that owns the AI? The person who programmed the AI? The person who trained the model? The artist(s) whose work was used to create the model? Or the person who came up with the prompts? For that, I don’t have an answer. Maybe all of them? Or none of them.
2: Credit. On the flip side, if the origination of the art and who gets to own it is a concern, then who gets the credit for even making it? And will the prompt writer just claim they created the work themselves? If they claim all the credit, are they lying about how the art was made? Does an artist who painted an oil painting need to give the canvas, brush, and paint makers due credit as well? Is that even a fair comparison? It has been a well-established standard that when an artist presents a work of art, they include the medium as well. Oil on canvas. Crayon on wood. Poop on a stick. Whatever it is, you include it in the description of the art. The same would likely apply here. Any art that was made using an AI should at least include “AI-Generated Imagery” (or some variation of that) in the description of the art, if not the actual AI tool used. In that way, you can at least solve the issue of giving the AI credit while also giving yourself credit.
The reality is that we have a lot to figure out still. Legally, are we going to treat this like a musician does when they sample another artist’s music? Do we include credit to every work of art or artist use to train the AI model? Or is simply listing the AI enough? Ethically, where do we draw that line? Are we free to use royalty-free and public-domain images? Or is using an AI at all an unethical practice?
Where the chips fall
It’s highly unlikely AI art will be treated like sampling a song since it is impossible to pinpoint what the AI used, where, and how. It would be next to impossible to ask an artist to credit every work of art or artist used to generate any one specific image, since the weights of the model would differ drastically based on a multitude of factors, from the seed to the prompts. And proving a work of art was or was not used would be a legal and technical nightmare. Including the prompts you use would also only be marginally helpful. not only are there a variety of other factors to include, but your prompts may not include anything specific to any artist or style and yet the AI could still pull from that style.
I don’t know where the chips will fall.
Legally, it’s likely we may have to give credit to someone in some way, but the specifics are still unclear. Perhaps future AI iterations will include a list of all the sources it pulled from, but even that seems like a tall order, especially considering the astronomical cost of getting the rights from every artist. Even crediting the model is a difficult task since they are so easy to make, manipulate, merge, and alter. It may be that future models will have to specifically list all the sources and get permissions from those sources. But in that case, who has that responsibility and how much of the final art do they “own”?
Created with Stable Diffusion
Ethically, it depends on how we decide to look at it. Is this just a new way of using reference material? Is this more like sampling music? Or is this outright theft? Is it ok to take credit for something you may have spent hours on, but the final image is still generated by an AI? Or will people be more open to using AI images as long as they are not used to make money directly or deter money from other artists? Again, that is going to be hard to prove. Even the new AI-detecting AIs will only work for so long before people or computers find ways around them, and proving an image was made by an AI may become increasingly difficult, if not impossible.
Personally, I think that giving credit to the AI you used as the medium, at least in part, is the least, and possibly most, we can do when using such images. More importantly, we should reconsider how we decide to use these images. Most artists will use references from movies to pictures to clothing, none of which they were involved in. Yet, they don’t have to credit any of these sources because the final images don’t actually use any of their elements in them directly. AIs are much the same way. While they might not be as good at using references as we are, they’ve only been doing it for a few years while humans have been doing it since pre-historic times. They may process things in their own way, and we may have to recognize that as they get more advanced.
I plan to use AI imagery, but at least for now, I plan to leave it in the development and previsualization phase. If we can use these images as references, we can use these tools to our heart’s content without worry. But eventually, there will come a day when it will take the stage front and center, and it may happen sooner than we realize. When it does, we should be ready.
It seems to me that image-generating AIs are going to become increasingly popular, and increasingly difficult to police. At least from my perspective, this appears more like an exciting new tool that has a lot of things to be worked out, but a lot of people don’t agree with me on that.
Unfortunately, our AI overlords don’t care. Either we adapt, or we die.