Man simulates time travel thanks to Stable Diffusion image synthesis
Throughout December, a social media user known as Stelfie the Time Traveller has been crafting a time-hopping travelogue using generative AI. Thanks to Stable Diffusion and fine-tuning, an anonymous artist has created a fictional photorealistic character that he can insert into faux historical photographs set in different eras, such as ancient Egypt or the time of the dinosaurs.
Stable Diffusion is a deep learning image synthesis model that allows people to create fictional scenes using text descriptions called prompts. With an additional technique called Dreambooth, people can insert their own subject or character into scenes generated by Stable Diffusion. It can also be used to insert real people into fictional situations.
So far, “Stelfie” has taken historical selfies during the ice age (being chased by a woolly mammoth), in ancient Egypt (during the construction of the pyramids), in ancient Greece (with the Trojan Horse), hanging out with Leonardo da Vinci (while creating the Mona Lisa), in the old West, while running from a tyrannosaurus rex, and while sailing with Christopher Columbus.