There’s a longer piece I’m mulling over at the moment—a dissection of art history as an experience (and as mere fodder for the AI cannon/canon)—but here’s something brief in the meantime.
I’m still grappling with AI and how it reproduces analogue photography. Last time, I borrowed from Jean Painlevé’s aesthetics. Earlier, I wrote about Stable Diffusion’s efforts with Cy Twombly’s Polaroids and Cindy Sherman’s film stills. So, as it’s New Year’s Eve and I’m feeling frothy, here are a few images in the style of fashion photographer Paolo Roversi.
“To succeed, you must combine the light of the past with the light of the future.”
- Paolo Roversi
Roversi gained a new audience about a year ago, when he collaborated with Catherine, now Princess of Wales, on a series of portraits for her milestone 40th birthday (below, left). But he’s been a big name in fashion photography for decades. I first encountered his work as a teenager, when I began to explore photography as a potential career.
Roversi’s signature technique is achieved with long exposures made on large-format Polaroid film (the artist famously bought up as much of the stuff as he could before it was discontinued). Of his practice, Roversi once said, “My photography is more subtraction than addition… [it’s] a kind of absence.”
For me, the notion of subtraction—both in photography and AI text-to-image generation—has become a recurring theme in my work with AI. After all, Stable Diffusion is a latent diffusion model; trained with the goal of removing and “de-noising” successive iterations of an image until the AI recognizes that the process is complete. According to Wikipedia,
“The model generates images by iteratively denoising random noise until a configured number of steps have been reached, guided by the CLIP text encoder pretrained on concepts along with the attention mechanism, resulting in the desired image depicting a representation of the trained concept.”
Think of this as a Magic Eye illusion, except it’s not your eyes discerning some latent image buried deep inside a field of computer-generated noise, it’s AI doing that, and you get a cool JPEG you can download when it’s done.
So, in that vein, here’s Paolo Roversi’s subtractive technique, rendered by Stable Diffusion.
These images were generated with prompts including Roversi’s name, as well as descriptors like blue, celestial, and atmospheric. While they’re perhaps not totally faithful to Roversi’s film photographs, I like them. The light is gorgeous, especially in the top right image, which almost resembles a Lillian Bassman print. Certainly, the one on the bottom, at the centre, looks a lot like the cross-processed fashion photography that was so in vogue (pun intended) in the late 1990s.
I’m reminded of pre-Raphaelite paintings, Cecil Beaton portraits, and ‘90s fashion photography all in one, so I think Stable Diffusion did a pretty good job.
But: how soon will we see AI-generated fashion photographs in the pages of VOGUE? For now, I have my doubts. Fashion magazines document pre-existing items of clothing—clothes that have already been designed and displayed on a runway.
AI has a future in the fashion designer’s toolkit (if I was a designer, I’d be making liberal use of it), but until it’s trained to tell the difference between a pantsuit from Stella McCartney’s Fall 2002 RTW collection and a 2019 gown by Chanel, I think it’ll be some time before AI is called upon to document designers’ collections for the fashion pages.
Long live photography.