Not really sure how to describe this. Started as an experiment that I didn't think would work but turned out pretty cool. Trained fairly lightly on a very small dataset.
pen-people.safetensors