The idea for this project originated from the novel “Naked Lunch” by William S. Burroughs. As a vanguard of the Beat generation, it is by far the most provocative, explicit, surreal, trippy, and chopped-up book I ever read (although not the most incoherent; that prize goes to “Finnegan’s Wake”). Particularly, it contains the quote, “Confusion hath fuck his masterpiece,” which in turn is a reference to William Shakespeare in Macbeth, “Confusion now hath made his masterpiece.” Allegedly, the new William had a knack for reworking quotes by the old William. There are several more cases throughout the book, but I am not familiar enough with Shakespeare’s work to point them all out.
Anyway, my friend Abram, quite the literary American, and I were very amused by Burrough’s confusion quote and joked around that matching t-shirts would be adequate. Furthermore, Abram suggested using baboons as a motif because they frequently appear in the novel (34 times, to be precise). Only what to do when you are too lazy and unskilled to draw baboons yourself? You make use of the increasing capabilities of AI; specifically, of generative adversarial networks (GAN).
I made some first attempts in Craiyon, but the results were all in all underwhelming. Good for a laugh, but not good enough for an entire project—let alone matching t-shirts.
Luckily, DALL⋅E 2 was just released and much more impressive in terms of versatility and accuracy. I tried a lot of different art styles and “baboons that look like Shakespeare,” or whatever the GAN imagined that to be. Oil paintings quickly turned out to be my favorite because they can be just coarse enough to not see the details that the GAN messed up, baboon eyes in particular.
Soon, I used other quotes from the book that I found amusing, or fascinating, or simply did not comprehend because the novel is so cryptic—as long as it fit the baboon theme. I was a bit lenient when it came to matching quotes with images, but be assured there always was an intention behind it.
Not much later, I began using general quotes attributed to Burroughs. Eventually, quotes by Shakespeare, especially insults I photographed from a postcard that I saw and did not buy in Salisbury, UK.
Sometimes, I did not have to add any captions because the GAN came up with new words, often even inventing new letters. I call this language “Boonish” and turned it into a small cryptology project, creating a cipher that resembles a constructed language.
The images were created in only three sprees and then released to a dedicated Instagram account at about one post a day, usually containing four images. In hindsight, it might not have been smart to spread some quotes across multiple images of one post, forcing users to swipe right to read the entire quote. But some were simply too long and would have covered too much of the baboons.
As captions, I used the GAN prompts, although I decrypted them using my own method “Lexigraph,” which is fun because it replaces words with other words of the same length, taking them from an English dictionary. Therefore, the encrypted texts contain real words, making hill climbing attacks less efficient. It is not a safe method by any means, and encrypting the captions really had no purpose other than being cryptic.
After I had released my initially planned 200 posts, I stopped using the channel and uninstalled the Gram because even though I only posted once a day and had prepared all posts in advance, I felt that it got to me psychologically and became more of a burden than a leisure activity.
The bitter truth: my first posts had gathered more reactions than the last ones, making the project quite unsuccessful in terms of social media response. This is unsurprising, as the channel was created in a time when the entire internet was flooded with AI-generated images. I myself grew very tired of those.
In case you are interested, I decrypted some of the art styles that I used for my prompts: cartoon, vibrant portrait, close-up, art station, explosion of a nebula, propaganda poster, anime drawing, Pokémon, high-quality photo, regal and religious art style, underwater shot with water bubbles, extremely detailed, particulate, studio lighting, ancient Egyptian mural, ancient Roman mosaic, Nuremberg chronicle, Byzantine icon, early Christian icon, lavish gilded codex, detail from Bayeux tapestry, bestiary of the Middle Ages, international gothic painting, autochrome photo, surveillance footage, black and white, grainy low-res dashcam footage, Daguerreotype image, disposable camera photograph, double exposure, drone photography, fish-eye lens photo, harsh flash camera photo, Lomography, camera obscura, telephoto lens, photo montage, collage, pencil and watercolor drawing, detailed oil painting, detailed painting, brilliant and bright and glossy.
I may release another batch of baboons once the next major GAN revolution happens. Back then, one post cost me about 20 cents, not including the prompts that were brazenly flagged as content violations by OpenAI, something that happened frequently due to the very nature of Naked Lunch. But ever since I allowed myself a high-end GPU, there is no need to use DALL⋅E 2 anymore.
Therefore, I switched to Stable Diffusion, which is open-source and tendentially more impressive (once I trained my own baboon model).
Lastly, here is the link to the corresponding Instagram channel @williambaboon.
Leave a Reply