The Power of Art Directing AI
On June 1st, 2022, I received an email: “We’re excited to have you as an early tester in the Midjourney Beta!” A week has now passed in what feels like a couple of hours. The image above was generated as a result of me writing a sentence filled with various descriptors about a yellow tent and a nature-filled landscape.
As an artist, I’ve been cautiously curious about the application of AI learning to the creative industry. After finally dipping my toes into this pool by generating some prompts of my own, I was ready to swim. But seeing the results of literally anything I can imagine come to life, it’s hard to describe how I feel now. It’s basically like Bill Hader’s impression of Keith Morrison from Dateline. The delight in my curiosity can’t help but shine through at the horror of this powerful tool.
Let me take you through my journey using Midjourney and show you some of the more interesting results of what I’ve produced (so far). The way this app currently functions is by posting your prompts to a chat thread on Discord. It will generate a block of 4 thumbnails from your statement and when it’s done, you can choose whether or not you want to upscale an image or show another variation of one. Living close to a coast, my first idea was “a crowd of people at the beach dancing in a circle during a sunrise.” Here are the results from that first prompt.
It didn’t exactly nail what I envisioned, but you can begin to understand how the AI is trying to figure out my instructions. From there, I decided to get more specific and asked for “a pregnant female model floating in space in the style of Francis Bacon.” If you are familiar with that artist, rest assured the AI did not fail at showing me something frightful.
Next up, I played around with making visuals of a character my wife had developed for a book she wrote. She loves Gustav Klimt, so she was delighted to see his art style applied to her creation. I’d love to use a tool like this when it comes time design another book cover for her. And according to the copyright verbiage from Midjourney, “You’re pretty free to use the images in just about any way you want as long as it doesn’t hit one of our current two special-case restrictions.” Those limitations basically say you need to pay if you are using the images as an employee of a large company or selling them as an NFT.
After this first initial tinkering, I plunged full force into satisfying my imagination. I started coming up with prompts to mix and mash my favorite bits of entertainment culture. I wanted to create and post the results to Reddit, Tumblr and/or Twitter just to see what the internet would say. I love the work of HR Giger, so I had this AI show me a xenomorph riding a bicycle as if painted by Norman Rockwell.
Actually, I spent a lot of time with prompts involving a xenomorph and a few days later, I landed on the idea of having it make me a marble statue of a xeno in the style of Michelangelo. Here is one comment from a reddit user after I posted the image of that statue to a sub called LV426:
It’s haunting; Disturbing and beautiful on many levels. Beautiful work.
The xenomorph is an easy model for the AI to imagine. Its body is very dark and intricate. The viewer has a hard time figuring out what they heck they are looking at. Sounds like a perfect model for an AI art bot to use (if it’s trying to pass as human).
Next, I moved on to music. What would the AI make of some of my favorite album covers? What does it know about the icons of 90s grunge? I excitedly entered prompt after prompt to see what it would spit back. It showed me Nirvana album covers from rock magazines of a different timeline.
Later on, I ended up with an alternate album cover for Ween’s Chocolate and Cheese and then Sonic Youth’s Goo. I posted them both to their respective subreddits to see what the fans would think. Here’s a reaction from one of the posts:
This AI s#*t is really wild, and it’s progressing really fast. I think in less than 30 years it’s gonna be dreaming up art/music/movies/TV shows that surpass human creative ability, and it’s gonna make lots of people really squeamish when the machines start to know us better than we know ourselves.
After tapering off with ideas from the music world, I wanted to get a better, base-level understanding of this particular AI called Midjourney. How does it work? Is it simply scraping Google and just mashing up the results? I started simplifying my prompts. All my non-detailed statements seemed to come back with creepy results. This particular robot is not going out of its way to make this particular human feel safe about the future of AI technology. Despite the scary factor on this next batch of results, I was impressed at how creative it could get with the existential material I was throwing at it. The image shown below was generated from the simple prompt of “life after death.”
Seeing what looks like a human figure rising from a candle flame after a panel of a skeleton ghost gave me a spooky yet satisfying chill. This computer just showed me a creative answer to one of life’s biggest questions.
After a minor, mental freak out, I went back to just nerding out. Ok computer, show me my favorite pop culture in artistic styles from different periods of time. I made a Predator movie poster if it was designed by Frank Frazetta. Then I went a little Frazetta crazy. What if he made a Stephen King book cover? What if he created the artwork for his epic The Dark Tower series? Then, I created a new Star Wars character by Boris Vallejo. I had Vallejo recreate Conan the Barbarian as if he was a futuristic robot warrior. Now lets see Vallejo do King stuff! Now let’s see Super Mario Brothers in the Mad Max universe!
You get the idea. I was beginning to see how this technology would be useful (or threatening) to the world of conceptual artists. Here’s another quote reaction from one of my reddit posts:
Goddamn Midjourney is either going to put some conceptual artists out of work or make them into rockstars.
After a couple days of this, my brain was starting to get a bit numb, and eventually, Midjourney told me I was no longer an amateur. It bumped me out of the “newbie” group and told me to explore other threads, like “characters” or “environments.”
When Midjourney renders your prompts, you also see the prompts and results of everyone in your thread. Seeing the descriptors of other users gave me ideas on how to get more punch out of my final image. I started using terms like “metric etching, soft lighting, unreal engine, cinematic, HD 35mm photography.” After tweaking and perfecting some details in my writing, I created the Lost City of Atlantis:
Over time, my results got better and more nuanced. After awhile, it felt like I was looking at super-detailed video game environments or amazing CGI movie backdrops. And once again, I could see the usefulness of this tool for artists. What if you were trying to get your film financed, but you were running out of time? Getting “your vision” across to a busy producer just got insanely easier.
Ultimately, my interest kept returning to the eerie creativity of these AI machines. Does this ability reveal an entity that understands the human experience? Are we witnessing the death of the human artist, or has technology given artists a new powerful tool in their arsenal? What would Leonardo Da Vinci think about all this for God’s sake?! Actually, what if Leo was shown some of our technology? And what if he witnessed modern military in action? And then he tried to sketch an F15 fighter jet onto paper. Let’s see what that looks like, shall we?
As you can imagine, the computing power to make all this happen costs money, and Midjourney cut me off after I had created about 100 or so images. They wanted $10 a month for 200 more images. Pssh… I quickly agreed. Sadly, I burned through those WAY faster than I thought I would, so I am cutting myself off for now. But, I would highly recommend you give their beta a whirl. According to my basic membership, I have 3 friend passes to give away and I would love to give you one. As soon as I figure out how.
Writing this article and not bombarding you with the images I created has been a lesson in restraint. Seeing the details of what an AI art machine can create with faces of humans and animals and geometric symbols has been astounding. It’s almost hard to grasp the amazing potential for machines like these to create seemingly endless variations of art. Witnessing the photo realistic nature scenes Midjourney can produce makes you feel like you have discovered a real deus ex machina. It leaves you feeling powerful, but taken aback.
Towards the end of my experimentation, I typed in just my full name as a prompt and it started to show me an image of a person. When the image resolved and sharpened to reveal a character that looked nothing like me, I honestly felt relieved. But I made it my Medium profile pic.
Cory Pinter is business analyst with years of UX design experience. He is open to work. Check out his LinkedIn here.
