Photo from Tom Betthauser, design by Mark Zimmerman
Astronaut Horse — an AI art exhibit with a Stable Diffusion AI that was trained on work contributed to this project by fellow artists — was held from Aug. 28 to Nov. 3 in UAA’s Kimura Gallery. TNL sat down with him over video-conference in late October to discuss his professional journey and outlook on AI art.
Tom Betthauser’s life has revolved around art from the very beginning.
Born in San Francisco in 1987 to a designer father and an illustrator mother, Betthauser was immersed from an early age in the rich artistic heritage of the Bay Area. While he was coming of age, another wave of enthusiasm — this time in tech — would sweep through his community.
“I remember my Dad was working in San Jose,” he said of joining his father on work-trips to the under-construction Tech Museum, “I remember feeling like there was a very positive, democratic general energy to that community.”
However, balancing the wildly differing cultures of artists and Bay Area tech entrepreneurs became difficult later on. As a result — from 2007-2012 — his studies focused sharply on art, extending from undergrad studies at the San Francisco Art Institute to a Master of Arts at Yale.
As the decade progressed, financial struggles led him to reconnect with the tech sphere he’d left behind in his hometown.
His old friends – many who are now software engineers – made him feel “re-inspired by this very small subset, or subsets, of that community.” The art classes he taught in San Jose slowly became more digital and web-focused, and by 2019 he was attending courses in full-stack software development.
Betthauser had some ideas for developing digital artists’ tools and he was admitted to a four-month software development residency at the Recurse Center in New York City in September 2022. Here, the story of Astronaut Horse would truly begin.
“There were a lot of people at the Residency. Talking about GPT-2 transitioning into GPT-3. In the process of our meetings, Chat GPT had become a thing. Prior to that stable diffusion.”
Stable Diffusion is an open-source — meaning code is accessible and viewable by the public — text-to-image generator. Compared to the closed-source DALL-E or Midjourney, it was far easier to peer into the models and training data the app used. It could be picked apart and reconfigured in innumerable ways.
Betthauser says the process of diffusion — the concept behind the tool he used for the art exhibit — “Takes a completely randomized static image and asks each pixel, in a very complicated calculus-related way: Hey, if you got darker, layer, more red, more green, more blue –would this whole image look more or less like this crazy math thing that I know is a fish?”
Betthauser felt this approach fit his needs more than DALL-E or Midjourney’s models — where two training AIs trick each other into properly understanding an image — because it required less image data overall. This meant he could use smaller bodies of work to create large collections of high-quality works.
He had concerns after his first attempt at using Stable Diffusion. Without his knowledge, it used his own artwork as training data, identifying the results as “Art that [Betthauser and his partner] had on our websites in the early 2000s.”
“I don't know if we were offended, but it felt like there was a strange intrusiveness to that. ”
After this experiment, Betthauser and a group of visual artists, students and software engineers began using artwork explicitly contributed by the artists for a new research project. This collaboration was dubbed ‘Astronaut Horse’ after a famous image generation benchmark for text-to-image tools: asking an AI to generate a picture of an astronaut riding a horse.
Betthauser expected that creating this project would take at least a year, but – with only a few weeks notice – his show was put on the calendar in Sacramento, and in Anchorage soon after.
When viewed in-person, the exhibition was a stark clash of simple composition and grandiose scale. Dozens of photo-printed images floated frameless along the wall, spanning the gamut of various artists’ work. Different genres and mediums were represented, with multiple pieces serialized to reflect Stable Diffusion’s ability to make many iterative changes on one image. Accompanying the still images was a video of self-transforming AI artworks in the back of the Kimura gallery. All were set to AI-composed music and represented several distinct styles as the images morphed into one another.
Astronaut Horse, according to Betthauser, acts as an exploration of “human relationships.” Betthauser said that AI art facilitates these relationships. “Even though those relationships are very different in terms of structure and scale than, I think, anything else.”
The physical exhibitions included training workshops for artists to interact with Stable Diffusion and discuss how they could use it as a tool for self-critique or workflow acceleration.
Betthauser expressed enthusiasm at this project and the future of others like it, but he emphasized a core belief that — despite the fact that tools can make it easier for artists – humans will always be the driving force of visual art.
“I have been consistently, positively surprised by all this. Art like this makes me wanna make art. It makes me more interested in exploring my own art. And it feels like it's hitting those notes without taking anything away from the human artists that I know.”
At the Strange Loop conference in St. Louis, Betthauser was asked at the end whether he thought AI art was indeed real art. In response, he recalled saying, “Oh, it is.”
While he believes there are still unanswered questions in the road ahead for AI art — particularly around intellectual property — he’s reconciled his conflicts between art and technological innovation.
Betthauser, now with his three-year-old at his desk, continues his work, illustrating and painting in analog formats while staking a claim in the digital world.
.jpg)
