June 27, 1970, 12:35 pm
I asked the machine to remember something it was never taught.
In June 1970, queer San Francisco marched up Polk Street. The next afternoon, about two hundred people gathered at Speedway Meadow in Golden Gate Park for a Gay-In. Now it is remembered as the city’s first Pride. Almost nobody was photographing the events. The record of those two days is thin.
I asked an Ai model to show me the crowd. The model sent back white men in leather and sunglasses, again and again. Faces melt at the jaw while hands come up short. And the skin tone holds itself inside one narrow range no matter what I type. Black, brown, transgender, and gender nonconforming people were on those streets. Here they are close to absent and nonexistent.
I started in 2024 with Analog Diffusion, a Stable Diffusion model trained on analog photographs for a film look. In 2026 I rebuilt the same settings in ComfyUI and generated roughly two thousand more images to see whether the first result would hold. It held. Same settings, two years apart, same crowd. The step count runs high and the faces still refuse to resolve, which puts the limit somewhere other than my process.
A lot of people in 1970 stayed out of photographs on purpose. Being queer or trans put your job, your family and your freedom at risk. Ten years later AIDS arrived. Many of the people in this crowd died, and their photographs and letters went with them. Whatever survived became training data.
When I look at these images, I see both things at once. People who hid to stay safe. People we lost. The machine sees neither. A thin record trains a thin model, and the thinness reads as neutral, because nothing in the image points at what is missing.
Whose faces did the record keep? Who stood outside the frame the whole time?
Composite, June 1970 is 32 Ai generated works. Every piece is a 1/1. A printed zine booklet, limited to 100 hand numbered copies, carries its own unique images.