Work

A diffusion model trained from scratch on 300 photographs from one film set, and the membership inference attack that finds them again. Scroll to watch it learn, then run the attack yourself.