In the MidJourney of Our Life

On a recent trip to the west coast I spent an afternoon immersed in one of my nephews’ latest obsessions, an app called DALL-E. It’s a program that uses artificial intelligence to (fairly instantly) generate works of art suggested by a simple caption written by the user.

For example, if the user types the prompt, “an armchair in the shape of an avocado,” they are rewarded with several images like this:

Avocado chair.

The AI scours the internet in search of inspiration, and reconstitutes its findings into the images you see here. The prompt “an illustration of a baby daikon radish in a tutu walking a dog” generates the following images.

Walking daikon.

You can imagine how a couple of teenaged boys who are addicted to pop culture might use this immediate-gratification brain to conjure up images juxtaposing Jedis and Jesus, Captain Marvel and pizza, and so on. I don’t have access to any of the images I saw on my nephew’s phone that afternoon, but from that brief crash course I had the impression that the software had a long way to go. Most of the human faces it generated were…off. Even the images we asked it to generate—of pop culture icons such as R2D2, Ernie & Bert, Spider-Man—were suggestive of the thing being asked for, not flawlessly clear.

So that was my experience of the hot new thing all the cool kids were playing with about a month ago. The thing could barely draw. If it got lucky, it would turn out a nice picture of a radish. Cute, but so what?

Cut to this week, when a professional book cover designer I follow and whom I have hired to create covers of some of my self-pubbed titles sent around a newsletter in which he sang the praises of MidJourney, yet another AI art app that also debuted about a month ago. When the designer typed the following prompt: “jack the ripper hiding in a smog filled london alleyway with red eyes,” the AI thought a moment before generating the following four images.

I think we can agree this is pretty stunning. And when the designer asked the AI for variations, he got increasingly stunning images of an atmospheric landscape.

Again, this is a machine surfing the internet, gleaning from it how things look, and then spinning off tweaks of its findings as quickly as possible in a new form. Compare the Ripper images above to, say, a the cover of an old issue of The Strand Magazine, or the cover of a book of Sherlockian pastiches. 

Both of these covers were taken from works of art created by humans and licensed by the publishers. The MidJourney images were made by software that can be yours for $10 a month, or $30 a month, depending on the number of images you think you’d like to create.

As a writer, I have zero need for such software, but I can absolutely understand why the book cover designer who sent these around to his clients was excited about this brave new world of art. Most designers I know lament how hard it is to find usable, abundant, or appropriate images on stock photo agency sites. It takes an endless amount of clicking page after page to assemble, say, five good images, from which you knit together a composite image and layer in enough special effects via Photoshop or Illustrator to make it look beautiful. That can be hours of work, for which you can charge an author, perhaps $700-$800, tops, for the resulting cover image. Imagine instead being able to type, click, create, and tweak an image in seconds—and charge your client for the AI’s results.

My last sentence is a bit of a fantasy. The designer told his followers that, actually, it still takes him a long while to coax the appropriate image from the machine—as long as it does when he’s searching stock agencies for photos—but what the AI produces is wholly original, one-of-a-kind art.

From the handful of articles I’ve read, people are already up in arms, in the same way they were when programmers started teaching AI programs to write like human writers. Some of the criticisms I’ve heard:

• These art programs are racist and/or sexist. If you don’t specify the type of person you want them to generate, inevitably they’ll return a white male. When women are depicted, it’s often in demeaning or violent settings. (Software engineers say they are correcting this tendency.)
• This means the end of decent paying careers for artists! (I don’t think so; not just yet.)
• If you write, draw, or create science fiction, fantasy, and horror, you’ve come to the right place. The rest of the creative world has no need for this type of crap. (If you have a strong stomach, you can check out some images intended for horror fiction here.)
• If indeed the AI is “borrowing” references from other images online, will it not inevitably infringe upon the work of a living artist? (I suspect not, but the monkeys have only begun to bang away on the typewriters.)
• An art director who hires many artists to create book covers recently waded into the fray, reassuring artists that they are not being replaced, and pointing out that the US Copyright Office does not permit art to be copyrighted unless it was created by a human being. You will probably already know this if you remember that story a few years ago about a photographer who wanted to copyright a photo that was taken by, yes, a monkey who stole, operated, then abandoned the photog’s camera.
• Already, AI work has won art awards, and human artists are pissed.

So, in other words, we are once again immersed in a kind of moral panic over something very smart people have dreamed up. I think it probably can’t be said enough that if a computer program is sexist, racist, or whatever, and it’s using as its creative input the freaking Internet, then we are truly reaping what we have sown. To the extent that images look lovely or appealing to our eye, that too is on us as a species.

That all said, I created a free MidJourney account and had a look around the communal chatroom where the bot is fed its prompts and where users can download the results of their queries. (You’re allowed 25 free uses before you are asked to subscribe.) I could immediately see that I was in the virtual presence of users who had a distinctly artistic mind. With experience, many of these users had come to know exactly how to word their requests. To get the bot to spit out an image in a particular style, they used buzzwords such as “unreal” or “hyperrealistic” or referenced specific artists or designers. Here are two highly specific prompts I encountered:

• Ethereal humanoid sphinx, Art by Peter Mohrbacher + h.r. giger + Zdzisław Beksiński + Tsutomu Nihei , unreal engine render , intricate details , hyper details —stylize 5000

• post-apocalyptic cebu city unreal engine cinematic digital art

Little ol’ unartistic me stepped up to the plate, and fed the AI the following prompt: “a detective named sleuth sayers who says sleuthy things.” Within seconds, I had these (admittedly male-seeming) images:

When I asked the AI to generate variations  on this theme, it returned the following:

Was it something I said?

* * *

See you in three weeks!


Joe
josephdagnese.com