Masked Diffusion Models (MDMs) have emerged as a powerful generative modeling technique. Despite their remarkable results, they typically suffer from slow inference with several steps. In this paper, we propose Di[M]O, a novel approach that distills masked diffusion models into a one-step generator. Di[M]O addresses two key challenges: (1) the intractability of using intermediate-step information for one-step generation, which we solve through token-level distribution matching that optimizes model output logits by an 'on-policy framework' with the help of an auxiliary model; and (2) the lack of entropy in the initial distribution, which we address through a token initialization strategy that injects randomness while maintaining similarity to teacher training distribution. We show Di[M]O's effectiveness on both class-conditional and text-conditional image generation, impressively achieving performance competitive to multi-step teacher outputs while drastically reducing inference time. To our knowledge, we are the first to successfully achieve one-step distillation of masked diffusion models and the first to apply discrete distillation to text-to-image generation, opening new paths for efficient generative modeling.
@article{arxiv.2503.15457,
title = {Di$\mathtt{[M]}$O: Distilling Masked Diffusion Models into One-step Generator},
author = {Yuanzhi Zhu and Xi Wang and Stéphane Lathuilière and Vicky Kalogeiton},
journal= {arXiv preprint arXiv:2503.15457},
year = {2025}
}