Sergey Prokudin-Gorsky (1863-1944) was a Russian chemist who was an early innovator in color photography. He studied under Adolf Miethe (co-inventor of the camera flash) to learn the early arts of color photography. In 1908, he gained fame for making Russia's first color photographic portrait, of Leo Tolstoy. This and other photos gained the attention of Tsar Nicholas II, who gave Prokudin-Gorsky funding to spend several years documenting the Russian Empire, using a special darkroom installed inside a railroad car.

Prokudin-Gorsky's methodology, based on Miethe's, involved exposing three sheets of glass covered in a light-sensitive emulsion, each recording a red, green, or blue "channel" of light. Most of his negatives were seized by Soviet officials after the revolution or otherwise lost. After Prokudin-Gorsky's death in Paris in 1944, just over 1,900 fragile negatives remained; the U.S. Library of Congress purchased them from his heirs in 1948 for about $5,000.

Assembling the negatives into prints was a delicate, complex, and time-consuming process — until digital technology made it much faster. In 2001, the Library of Congress put on an exhibit titled The Empire That Was Russia: The Prokudin-Gorskii Photographic Record Recreated, bringing many of his photos to light for the first time. They were something of a sensation, offering an eerily contemporary-seeming view into life on the Russian steppe a century earlier, a time we're used to seeing only in black and white.

This embedding is based on 30 images of Prokudin-Gorsky, focusing on his work with people as opposed to his many landscapes. I confess that I was hoping this would capture some of the glorious color misalignments that populate his color prints, giving them a spooky, ethereal quality. (Part of that is due to the fact the red, green, or blue images would be taken seconds or even minutes apart.)

Alas, Stable Diffusion was too smart for that. So instead, it captures the rural Russia that Prokudin-Gorsky recorded, full of peasants, farm workers, and other people from the many minority cultures he captured, including Dagestani, Bukharan, Jewish, Armenian, Bashkir, Kyrgyz, Karelian, and more. It was cooked for a total of 300 steps on base SD 1.5: 16 vectors per token, a 0.004 learning rate, a batch size of 6, and 5 gradient steps.

Your ratings, especially of the five-star variety, are very much appreciated.