Well, GAN based image generation is also just one step and fast-fast. But, diffusion models due to them working in increments are way easier to train and fine-tune. That what actually made them a dominant solution. Also, they can be optimised, including to work in just four steps.
A team at MIT has shortened the multi-stage processes of well-known AI image generators. This not only reduces the time it takes to produce the finished image. The required computing power and power consumption are also reduced at the same rate.