We develop flow map methods for state-of-the-art few-step generation, generalizing flow, diffusion, and consistency models.

We introduce a framework for training accelerated, few-step generative models that includes consistency models, shortcut models, and mean flow as special cases.