AI ArtL: How It Does What It Does

Interestingly, one of the major issues with AI art is that the technology has advanced to such a point that some of those most affected by it don't know how it works. Too many artists are a little behind the curve when it comes to the technology and it shows in the discussion regarding AI art; as such, it puts them at a disadvantage when it comes to discussing the potential effects of the technology as well as its legal ramifications. In order to get up to speed, it may not be a bad idea to take a step back and look at what it does.

The Quickie Version

At one point, "AI art" was a serious misnomer. While the software could modify and merge images, it needed a human operator to set the parameters; you could flip, do some simple animations and morphs, and even simulate living animals based on photographs, but it needed a human at the helm. However, artificial intelligence would advance, and as it did so, so would its capabilities. While it still needs some human guidance, artificial intelligence has advanced to the point where it can be trained, create some sort of art by itself, and help artists work out some issues.

However, this is just the beginning of the process. An actual artist needs to take the images so created and modify them, compose them, and even put the final touches on them. This means that the human artist needs to initiate the process and complete it; artificial intelligence is just not capable of creating true art on its own and is a ways away from doing it just yet.

Let's Start With The Prompting

For our purposes, the first step is someone entering some sort of prompt for the software to act on. The software requires a prompt in order to figure out what to do. The more precise these prompts, the more likely the software will be able to create the art required by its operator. Ironically, the best prompt designers are actual artists, especially those with a background in art history, pop culture, and even art appreciation; they can design the most precise prompts with the least iterations possible.

It needs to be noted that artificial intelligence generally requires several attempts ("iterations") to get to the best possible choice. This is why the prompts need to be as precise as possible; the more exacting they are, the closer the artificial intelligence will get to the desired result, taking fewer iterations. Only if the person doesn't really care about the results does this take just one or two tries; it usually takes ten or more tries.

It should be noted that the prompts are limited by the amount of detail required. Consider Spider-man's trademark logo; you would need to be very precise in order to copy the spider, especially in terms of species, size, facing, and even position. Now imagine doing the same for his entire costume. Ready to do a scene with multiple superheroes in it now? There is a way to do it, but the important point here is that the prompt needs to be kept as simple as possible while also being as precise as possible.

Then There Is The Training

This is the "scraping" you keep hearing about: The AI is either given specific images to consider or is let loose on the internet. Either way, the AI needs to convert it into a set of instructions so that it can replicate the style of the artist requested. The key in that sentence is "set of instructions"; the software needs to be able to replicate the style as much as it can possibly using subject matter that the original artist never considered. If it had to rely on just shifting bits around then there is no way that it would be able to handle some of the prompts it was given, such as "Batman in the style of van Gogh".

There are three things to keep in mind: First, it's boiling the images down to more instructions. This means that while it may record the images used and even note their keywords for later reference, it is essentially throwing the image itself out; it's just not necessary once it's been boiled down to its instructions. Second, this is where the "unreadable signatures" are coming from: The software is noting that some sort of squiggly line is part of the style and so it's included. It's worth noting that this will change from iteration to iteration; it won't look the same every time.

Third, this is how most artists learn how to draw, especially if they are trying to copy someone's style on purpose. Yes, you can quibble that the specific method is different, but the bottom line is that most artists will study another artist's style, break it down to its simplest concepts, and then reconstruct the style from those notes. So while the specific mechanics are different, the artificial intelligence is, just like its human counterpart, taking notes and applying them when it does the actual piece. In this way, both intelligences, human and artificial, are on the same page.

Putting It Together

This is where things depart from some assumptions in a big way. Some people assume that the AI is pulling the original image(s) apart and putting it(them) together in a new way, like you would for a collage, just a digital version. For some programs, they would be right; originally, a lot of programs that morphed images did exactly that. However, the current software doesn't: it creates the piece of art from its "notes" whole cloth. The piece of art created is therefore wholly original; the image has never been seen before and has only been made at the moment.

In essence, rather than taking several cakes and pulling parts of them together in order to create another cake, the software is grabbing fresh ingredients and attempting to create a cake in the style of a known chef.

While this has some interesting ramifications, the point is that whatever is created is not a mere collage; it's a fresh creation. These ramifications will be explored in a different article, but for now, it's only necessary to recognize that the piece created is NOT a collage, but something new.

But…Is It Art?

If the person stops right here, then it could be honestly debated that the piece is not art: It lacks any will or intent and therefore is not art, which requires will and/or intent. So far, the various groups who rule on copyright matters have agreed and have either ruled that there is no copyright or that the holder of the copyright is the person who initiated this process.

However, you'll note that this assumes that the person stops as soon as something has been spit out. For some, this is just the beginning of the process. The person will modify the image to deal with certain issues, such as the hands, small details, and any number of color and texture problems. They may decide to take parts of various iterations, taking the best parts of each to create a better image. They may even create specific iterations for specific functions (such as the main figure, the clothing, and weapons all being from different prompts). While this composed art does resemble a collage, it can also be argued, because of the will and intent expressed by the artist that this is actual art.

So, What Does This Mean For Artists?

All told, this isn't going to mean much. While AI art is going to be a boon to small studios and independent non-illustrator artists, and some companies may use it for some of their smaller projects (including concept art), in general, most artists won't notice the effects of AI art: Especially at this stage, the limitations of the software require a digital artist to make the most of the software. In fact, this may even make the life of an illustrator even easier: You know those irritating people who want you to draw for free or at reduced rates? You now have a place to send them.

In short, while some illustrator jobs may be lost, they should be relatively limited and you may actually see an increase in digital artist jobs; some companies may even be looking for artists to create prompts for them. Also, a lot of the people using the software would be bugging you for free work anyway, so you may see a minor decrease in commissions but a major decrease in aggravation.

For those arguing that the cost is going to down, the reality is that you still need to pay for the time and skills of an artist to come in and do the heavy lifting; that part of the equation won't be changing any time soon.

Regardless, you shouldn't be worried about AI art. Between its limitations and the learning curve of those using it, it's more problematic and more expensive for companies looking for a good quality product. While the technology doesn't work quite the way some think it does, it may actually help the art industry more than harm it.