DM: What about the use of stock imagery, and how the figures in these images are digitally “deleted” or removed? Is this done using software, or is it “hand-done?”
KM: I do the wiping by hand, using the “correct” feature of one of those photo editing apps you can download to your phone for $5. I had initially wanted to use the artist damjanski‘s app, Bye Bye Camera, to create some quick imagery to go along with the text—he wryly calls his app “posthuman,” and that seemed to fit conceptually the idea of loss. But I found that the industrial apps give me a lot more control over the results. They use a simple form of machine learning to fill in a background when you ask it to remove a figure. For example, people typically use the feature to remove someone photobombing their perfect sunset snapshot. The delight of this bespoke process has been in “misusing” the function for my own ends.
Initially my strategy was to look for stock images where, if you remove the person, there is still an indication of human presence. You don’t want the wiping to be entirely successful, otherwise it just looks like a still life. If you remove the person and leave the shadow, there is a trace, a human presence in the photo, still. So, the earliest images I worked with featured bicycles in mid-ride with the rider removed, or umbrellas that seemed to be holding themselves up, or a skateboard seeming to flip in mid-air by itself. But as the project has progressed, I have moved more toward only partially wiping the figures. So: a face without eyes, or a hand disconnected from its body. Turns out you only need to remove a small part of a person to interfere with their humanity and agency. I think this evolution came from the experience of writing the text, and realizing how few body parts I can actually refer to. If I can’t use the letter “E,” I can’t say: eyes, face, head, feet, legs, heart. What happens to those body parts if they can’t be named? Do they still exist? How do you move through the world without language?
I generate the short videos by “wiping” the same area several times. The algorithms don’t create the same-pixel-by-pixel result each time, so there’s a bit of variation in each image. Taken as a composite video, the images look like they’re shimmering.
View this post on Instagram
DM: The language and captions in the posts are also synthetically generated without the use of the letter “E” — why “E”?
KM: I might have picked a different vowel to omit. I considered omitting “I” because it seemed interesting to take up the impossible task of writing a first-person narrative in which I can’t refer to my own person. But I chose to omit “E” because in English, all regular past tense verbs end in “ed.” If you remove the past tense, time warps in strange ways. My narrator moves back and forth fluidly from a very constricted past tense made of only irregular verbs, to a present tense meant to be understood as the past. It was a way to poetically evoke the unreal way we experience time without history during lockdown.
DM: How is the language generated? Do you take it as-is, or do you personally edit or control what you decide to use?
KM: I use an algorithmic GPT-2 type text generator that has been modified to omit whatever letter I want, in this case “E.” It produces coherent sentences, but they are very bland. Sometimes I’ll pick out a few phrases verbatim, but there is a lot of me really composing off those bland sentences, adding rich bits of vocabulary, thinking through the storyline and character development. Injecting, you know, humanity and imagination. The process is a lot about me assembling language, ideas, and a sense of the mood I want to evoke in any given post—and then riffing off the elements I have gathered at hand that day.
An algorithm can never output anything as coherent as Perec’s novel, A Void. But it is a handy tool for amassing a lot of language that is constrained in a certain way, so that you have something to build from. Plus, as coherent as A Void is, I find it largely unreadable. The pleasure is conceptual, in knowing Perec pulled off the feat. For Voidopolis, I work really hard to make each post not only a technical feat, but also vivid and muscular as story. I want people to be captivated by my character Nikita, to want to read the next episode. The short format of Instagram captions helps a lot: I don’t think I could sustain the same level of interest and excitement if I was composing an entire novel in this mode.