Models that create rather than classify — autoencoders, GANs and diffusion — and what it actually takes to put a deep learning model in front of real users.