Diffusion models make great images and videos but are slow because they usually need many tiny steps.
Diffusion models make pictures from noise but often miss what people actually want in the prompt or what looks good to humans.