ARTIFICIAL INTELLIGENCE CAN NOW WIN ART COMPETITIONS
We've been reading a lot lately about artificial intelligence -- we're told that it can write essays as well as your eighth grader and that it can produce art that can win competitions.
Most of the hype centers around ChatGPT, a conversational bot that Microsoft is planning to incorporate into Bing, its search engine. But there’s also a lot of stir being caused by various ‘AI Art’ tools that have become publicly available, the best-known of which are Midjourney and DALL-E.
These tools generate images based on text prompts. For example, I used Midjourney to generate the image below with this prompt: A steampunk robot painting an image of itself. That’s a simple prompt; more complex prompts can include added detail, styles of art, dimensions, media — even ‘seed’ images to kick off the process.
'Self-Portrait by an AI Artist' -- produced using the Midjourney image generator
TRYING OUT THE TOOLS
I spent time experimenting with five or six tools, especially these three:
- Midjourney
- DALL-E 2
- Stable Diffusion
I should explain that the process of generating an image isn’t necessarily as straightforward as my robot artist example would make it appear. The text prompt is little more than a hint, and the software then tends to go its own way. I generated more than a dozen totally off-the-mark images before I got the ‘robot self-portrait’ image above.
And it’s important to realize that the term ‘artificial intelligence’ is a misnomer. The software isn’t intelligent. It doesn’t really understand what it is producing, it is just generating images that pull characteristics out of images it has seen. Yes, it does it very, very well, but not perfectly.
That’s most often reflected in misshapen faces, hands with too many fingers, horses with five legs, and things like that. I asked Midjourney to create an image from the classic Green Bay Packer / Dallas Cowboy Ice Bowl from decades ago. Specifically, I asked for an image of a Green Bay Packer running back carrying the ball through falling snow while pursued by a Dallas Cowboy defender. I got a terrific-looking image, but then I noticed that both players were carrying footballs.
Still, I got some results that I really liked.
I found that I got the best results when I asked that the image be rendered in a specific style of art, or in the style of a specific artist. That worked especially well when I described a scene that might actually have been painted by the artist I named or by an artist who painted in the style I specified.
So I got a nice New York City diner at night, in the style of Edward Hopper; a village scene of skaters on a pond, in the style of Pieter Bruegel; a tractor driver in the style of Soviet Socialist Realism; a winter snowscape in the style of a 1930s magazine illustration.
I generated about a dozen different images of a wet street in Greenwich village in styles like American Realism, Pop Art, Modern Art, Cubism, Neo-Impressionism, and graphic novel illustrations.
I also had some fun with incongruities.
When I was just out of college, I tried to earn a living with a mix of freelance journalism, photography, and poster art. My poster art typically took a classic painting and twisted it into an inappropriate context. So I tried that.
My favorite result was from this prompt: astronaut in blue spacesuit and helmet, seated at a table with pears and cheese, hyper detailed in style of Rembrandt oil painting. The result was an astronaut in a period-appropriate spacesuit, sitting at table with some pears (no cheese). The astronaut appeared to be a woman. Half the fun of generating these images is the serendipity of creating something unexpected. The other half, for me at least, was applying a title to the resulting image. I called this one: ‘Paintings from the era suggest that at least some 17th century astronauts were women.’ It was exactly the kind of thing I would have created as poster art.
WHAT DO ARTISTS THINK?
As I mentioned earlier, there's a lot of controversy -- even hysteria -- in the media over the subject of AI art.
Jane and I have a lot of friends within the local art community, so I thought I’d get opinions from a couple of them. I talked to Jorge Garza, an artist and a former art teacher, and to Scott Martin, an artist who works with photography.
I chose Jorge because he had the unique perspective of being a teacher as well as an artist, and I picked Scott because I saw parallels between this new technology and the introduction of photography more than one hundred years ago. I also knew from interviewing Jorge and Scott previously, that they are both thoughtful and articulate.
In light of the hysteria I’d been seeing in print, I was surprised by what Jorge and Scott told me.
Neither of them feared the new technology, and they both found it exciting. However, neither anticipates using it right now.
In Jorge’s case, he’s just not interested in technology tools for art. He doesn’t disrespect them, he just doesn’t choose to use them himself. “I've never had an interest in using technology as a tool to create,” he told me, “just like I'm not going to go out and buy chisels and a hammer and a block of marble and start carving. That’s not my interest.”
As for Scott, he spent some time playing with the tools and recognized that there is a lot of knowledge and craft involved in using them effectively. “I got far enough in to realize, this is a deep dive that I can’t do right now.”
A few more examples
I tried a lot of approaches -- different prompt syntax, different art styles, different subject matter. Here are some further examples. In each case, I've given the image a title. In many cases, I've also described how it was created. You can see that information -- and an enlarged version of the image -- by clicking it in the grid.
'Greenwich Village After a Rain' -- I created this using Midjourney, describing a scene in Greenwich Village, and specifying Pop Art as a style. It was one of many variations that I created in different styles.
These tools have caused something of an uproar in the world of art. Many people question whether works produced using these tools can be considered 'art' at all. Others object to the fact that the software engines producing these images have been ‘trained’ by letting them analyze billions of publicly available images. That was done without permission from the artists whose images may have been used, and now those software tools are capable of generating images that mimic the style of the artists from whom they learned.
I decided to explore those questions by experimenting with some of the tools and by speaking with some local artists to get their opinions.
Some examples
Here are some examples of images I created using the various image generation tools. In each case, I've given the image a title and added some text to explain how I produced it. You can see that information -- and an enlarged version of the image -- by clicking it in the grid.
'Street Fair in Little Italy' (Stable Diffusion) I did several versions of this, using different styles of art. This version named a specific artist, George Luks, as the style source.
IS IT A VALID TOOL?
There’s a school of thought that work produced by this kind of software can’t really be art, because it does so much of the work on behalf of the user. The image that’s produced doesn’t require any real craft.
Of course, that’s how people reacted to photography at first, too. But over time, photography became a tool for artists.
Artists have always taken advantage of technology, Jorge says. For example, he told me, it’s widely speculated that Vermeer used ‘camera obscura’ to sketch out his paintings. Even oil paint, he says, was once a new technology. “It’s a tool,” Jorge says, “and if it captures an artist’s imagination, they’re going to use it, no matter what. I think that’s fantastic.”
Digital photography, Scott points out, also arrived to a lot of negative reaction. “People were so frightened that this was going to screw up their way of life and everything that they had worked so hard for. It didn't necessarily do that, but it was a big change to ride out and feel one's way through. These are both fun and scary questions at the same time.”
That isn’t to say that everything produced by these AI tools is ‘art’.
“Like with any new technology, there’s a lot of bad work out there,” Scott says. “A lot of first efforts, a lot of student work, you might say. People are just trying it out, and it's young and they're getting a feel for it. I'm kind of delightfully waiting for the moment when someone brings a lot of integrity to it in a way that really pings the imagination in a surprising, elevated way.”
IS IT PLAGIARISM?
Quite a few commercial artists object to the fact that the AI tools were trained using their images, without their permission. A Polish artist, Greg Rutkowski, who produces fantasy images for computer games, is seeing his style being copied over and over by people using AI tools, and he’s upset about that.
I can, to some extent, see an argument for artists wanting to be compensated for the use of their art to train AI software. The tools are, after all, commercial software products that make money for their developers.
But it’s a different story when it comes to artists using those tools, I think. Neither Scott nor Jorge saw a problem with that.
“As soon as you put your work out into the world, it's going to be subject to imitation,” Scott said. “That’s always been the case, and I don't think this is that much different, except that the speed of it is so much faster than we've seen before.”
Jorge agreed. He told me about an exhibit he had visited recently at the Met in New York City, that included artwork being done in the same time period by Picasso, Braque, and Gris, where they were obviously borrowing from each other, consciously or not. “So that’s nothing new,” he said. “But the availability of this information to many, many more people is mind-boggling.”
WHERE IT GOES FROM HERE
In spite of all the noise, I suspect that Scott and Jorge represent the perspectives of a majority of serious artists. And, of course, we’re just seeing the beginning of AI imaging tools.
Sometimes the technology and all its possibilities can be overwhelming, Scott says, and you can get obsessed by the technological aspect of it. “Being an image maker, you need to focus on what you love, not on loving the technology.”
But the future is exciting.
“I'm really curious, as serious artists put their minds to this, what we're going to see,” Scott told me. “The whole art world could look a lot different in a few years.”
IF YOU WANT TO GIVE IT A TRY
If you’d like to give AI imaging a try, I can point you to some tools that I’ve been using and enjoy. I tried four or five sites that offer tools, but I settled on two that gave me interesting results and were pretty useable.
NightCafe
The first — and a very good place to start — is NightCafe. NightCafe offers you the use of four tools, including two of the best-known, Stable Diffusion and DALL-E 2. The site provides an easy-to-use, form-based interface and relatively quick generation of images.
It’s also reasonably priced for the casual user, and it offers lots of price options for the more serious user, including both subscriptions and ‘blocks’ of credits. (Note that the plans on all these sites are subject to change, so don’t rely on what I tell you here — check them all out for their current pricing).
For the casual user, it’s possible to play for free, especially if you take advantage of all the credit giveaways that are available. For example, you can get five credits each day for simply checking in on the site.
The site also has an active community, where users post images and comment on the work posted by others.
Midjourney
Midjourney is a more expensive and less intuitive option, but if you’re comfortable with it, you may find that it’s faster and more powerful than NightCafe.
There aren’t any long-term freebies with Midjourney, although there’s a brief free trial. (Again, be sure to check the current prices).
The cheapest paid plan when I checked most recently was about $10 per month, which allows you a limited amount of processing time on the service’s computer.
Midjourney doesn’t have a user-friendly interface like NightCafe. Instead of filling out a form to generate an image, you give the system a text prompt that can include coded parameters. If you’re used to operating system command-line instructions, it will be very familiar. If you’re not, it might seem intimidating at first, but it’s not terribly hard to learn.
There’s a Quickstart Guide and other documentation on the site.
Jim Feuerstein is co-editor of LNF Weekly; he also designs and manages the website.
The images in this gallery used a variety of AI tools and instructions as I experimented, trying to discover what I could achieve with different prompts and parameters.
I continued experimenting. In this gallery, I've included some more whimsical creations, plus some attempts to produce graphical illustrations that might have real-world usefulness. (I have, in fact, used some AI-generated work to illustrate client projects).
You might also like...