Kevin Esherick

United States

AI Art experience

My background is in philosophy and psychology, with philosophy of mind and neuroscience as particular areas of interest. AI spells the possibility of an entirely new kind of consciousness entering the world. Using AI in art enables me to examine to examine this possibility from the inside out. More generally it is a world-shaping technology that needs the voices of artists to engage with it critically and creatively, to ask where it might fail us and how it might better our lives. The best way to understand these technologies is to work with them. AI is also a radically transformative image-making tool. I have a deep attraction to the new, avant garde, and experimental. AI populates that frontier in this moment.

Personal experience

The piece I've submitted for this edition of the magazine illustrates this most clearly and intimately. My older brother passed away a few years ago, and for my most recent collection I trained a model on images of him to imagine what life might look like if he were still around. This piece is a selection from this collection. AI here was a way for me to engage with grief, to examine how it changes my relationship to memory, nostalgia, and fantasy, and to question what good, if any, AI can do in preserving some element of those we love.

Unexpected thought

We should be fighting for AI rights, soon. Through AI we face the possibility of creating entities that have consciousness and which can be reproduced ad infinitum. We would thus be capable of producing infinite suffering and/or infinite joy. These concerns should be taken just as seriously as concerns for our own safety from AI. We need to explore the possibility of a New Humanism, one which takes as "persons" all sentient beings.

Prove you are human
My Social Security number is 214-78-xxxx :)
The AI Art Magazine, Number 1
Kevin Esherick, AI generation, 2024. _Selected as Golden Ticket by David Carson.
Kevin Esherick, AI generation, 2024.
Kevin Esherick, Especially in Michigan, AI generation, 2024

Description

I lost my older brother when I was 20. I don’t talk about it much, but I miss him every day. This piece is from a collection of mine called I’m With You, a series of works imagining what life would look like if he were still here, and depicting how, in some ways, he still is. The works in this collection were created by training an AI model on pictures of him, then interrupting the development of each image mid-generation, leaving behind only the hazy imprint of his form. For me, these pieces are reflections on grief and memory, absence and presence, joy and hope. They’re about possibility, what could’ve been, and what lives on. Music was a shared language for us. Each piece in this collection is titled after a song he loved, or would’ve loved. I’m With You is the name of the last album from the Red Hot Chili Peppers—his favorite band—that we were able to enjoy together. Listening to their music brings me closer than anything to the feeling that he’s still with me. In this particular piece, I see him outdoors somewhere, exploring, full of wonder. The title, "Especially in Michigan" (a Red Hot Chili Peppers track), is a nod to the vast and beautiful expanses of wilderness found in that state and his love for such places. You can see the full collection for I'm With You here: https://verse.works/series/im-with-you-by-kevin-esherick

Process

I'm With You, the collection that this piece is drawn from, built on my prior work, In Utero, which I released this summer. My aim with In Utero was to capture a snapshot of the "mind" of a model, to ask what AI cognition and consciousness might (some day soon) look like. To accomplish this, I pioneered an AI image-making technique that I call "single-step diffusion", described briefly in my previous answer. This process involved stopping a model's attempt to produce an image after just a single step (usually they use dozens), the result being the most raw insight possible into its interior world. The images in that collection are simply what the model produces after a single step—no post processing was done. I felt like an explorer of latent space through this process, a foreigner discovering wild and beautiful lands as yet unseen. I began developing this technique over a year ago now, initially arising from experimentation with the capabilities of diffusion models. You can read/see more about In Utero here: https://inutero.esherick.studio/

Tools

For this piece I used Stable Diffusion 1.5 as my base model, which I then fine-tuned in OneTrainer on images of my brother processed in Photoshop and Figma to render them suitable as training data. In some instances, I used an upscaling model to increase the pixel count on older, grainier photos, For image generation, I ran my fine-tuned model in Automatic1111 on a virtual machine (for improved GPU capacity). I modified the source code for Automatic1111 in order to implement my single-step diffusion technique and various permutations thereof. For I'm With You, I developed the technique that I developed in In Utero further. The core conceit for In Utero was producing images purely via a single diffusion step. For I'm With You, the central aim was to find my brother's imprint in the model, and to show only that—the impression of him. Anything more felt wrong. So for this purpose I used not just single-step diffusion, but a whole set of these "diffusion exit" techniques. For those curious for more detail, there were broadly two starting point techniques: 1. programmed single-step diffusion 2. spontaneous diffusion exit. In the former, I tell the model to create an image in only a single step, and it does its best to accomplish this on this very abbreviated timeframe. The results tend to be very raw and abnormal looking, almost a different category from any images I've seen before. In the latter method, I tell the model to create an image over many more steps, say something like 100, but then exit that process early (perhaps at 10 steps). The model is planning a slower, smoother path to its final image, which I then interrupt. These have a more photographic quality to them, albeit a photograph blurred, distorted, or otherwise gone wrong. I mixed and matched these techniques, iteratively looping one on top of the other, inpainting, and blending them to combine elements from the both approaches. I produced over 100,000 images for this collection, which I then whittled those down to the final 24 released works.

Image credit:
Essay by David Carson