I futzed around with AI art and this is what happened
I’ve been playing around a bit with AI Art generators, and my thoughts about them have careened across both ends of the hot-take-spectrum. Anyone who’s tried them has probably had at least one moment of “holy shit,” but for a lot of artists (whether or not they’ve experimented) there’s nagging ethical questions.
That’s where I started, when I first learned these existed - how can this be ethical? These AI systems were trained with mountains of copyrighted data, and they are now using that copyrighted data as the basis from which they produce new images. Doesn’t seem cool, right? Who moved my (or Monet’s or Jeff Koons’ or Banksy’s) cheese?
But then, on the other hand, there’s the Holy Shit factor. How is this possible? Why would I (or David Hockney, or Robert McCurdy) ever try drawing or painting something again?
Anyway, I’m not prepared to dive too much into either of those perspectives. At this point, all I can do is experiment, think more thoughts, and see where it all leads me. I started with DALL E 2, and I have several more weeks of experience with it than Midjourney, but I’ve played with both, and will get into describing some of the differences.
For ages, I’ve been keeping a list of ideas for paintings. Sometimes I actually execute the ideas, but many of them just languish on the list. What a great starting point for an AI art generator, I thought! I began feeding DALL E some of my ideas, and was generally surprised and impressed with what I saw being kicked back.
There were absolutely some quirks and challenges - the AI struggled with faces, consistently horrifying me with the renderings. Body parts, especially limbs or digits, were often “incomplete.” But, depending on the prompt, some more impressionistic or abstract ideas came back with a looseness of technique and adherence to direction that felt astonishing.
Still, as impressive as the successful images were, they didn’t sit right with me, existing as they did. My lack of participation in their creation left me cold. My hands weren’t dirty, and if my hands don’t get dirty, it doesn’t feel like art to me.
So I decided to use the AI to iterate on ideas, until I found a palette and composition that I liked, and then use that as a reference for an actual, honest painting that I would make with my human hands and a brush and wet paint.
The AI’s take on a vase of roses and lillies.
After giving this process a try, I came away with a clearer sense of how I might continue using these tools in the future. The “painting” the robots made, and the painting I made from that “painting,” are different. Creating mine felt more like an act of creation, more like creativity, more like completing a work. It opened up more opportunities for serendipity, surprise, and expression.
There was a brief moment after generating the AI image when I felt like I made something, but painting the flowers myself was overflowing with those moments. Regardless of the results, one of these methods definitely felt better, and filled up that cup of mine, that I think all artists have, that’s always on the cusp of running dry if we aren’t regularly making stuff.
Knowing that I found a way to use these systems that felt decent, I was free to try another one, so I fired up Midjourney to learn what it could do, and where its results diverged from DALL E’s . Out of the gate, it clearly solved some of DALL E’s trouble with the human form. (Who hasn’t always wanted a Ghostbusters death metal record?)
I thought it would be fun to give both systems the same prompt and see what they came up with, so, I went with:
“combine the styles of piet mondrian and egon schiele to illustrate a colorful telephone of the 1980’s” This is what came back from DALL E:
And this from Midjourney:
At first glance, the gut reaction is how much “better” the Midjourney image is. But reflecting more, I realized that DALL E may have done a better job actually following the prompt.
Consider it: Mondrian and Schiele (to my knowledge) never made a two dimensional work that was intended to look as realistically three dimensional as the Midjourney image. Neither of them were in the business of photorealism, especially Mondrian. So when I tell these robots to replicate their style, and I get back something photorealistic… did the robot really listen? (Also… IS THIS THE BEGINNING OF THE SINGULARITY!?)
That said, the quality of the image, the creativity, the je ne sais quoi… is lacking in the DALL E image. It’s flat, like a Mondrian, although I don’t think I’ve ever seen that shade of magenta in his work. I don’t know what it captured of Schiele, if anything. But it doesn’t make me feel anything, I don’t have any questions about it, it’s forgettable.
The Midjourney image, however, has my mind reeling. This looks like an actual product. Like a sculpture. Like a representation of an idea. Like Jony Ive had been hanging out in Edison’s lab at the turn of the 20th century. I would paint this in a mural, on the side of a building.
Is it what I imagined when I fed the prompt? No, and that’s what’s so miraculous about it. I didn’t really imagine anything, and I don’t think I could have imagined this, but now that I’ve seen it, I can’t unsee it.
So what’s next?
I have no fucking idea.
But I’m going to keep futzing around until I find out.