On the Design of One-step Diffusion via Shortcutting Flow Paths

TL;DR

This paper introduces a unified framework for designing one-step diffusion models that shortcut the diffusion process, providing theoretical insights and practical improvements, leading to state-of-the-art results on ImageNet-256x256.

Contribution

It offers a systematic design framework for shortcut diffusion models, clarifying their components and enabling principled improvements without pre-training or distillation.

Findings

Achieved FID50k of 2.85 on ImageNet-256x256 with one-step generation.

Reduced training steps to 2x while maintaining high image quality.

Provided a theoretical foundation for shortcut diffusion model design.

Abstract

Recent advances in few-step diffusion models have demonstrated their efficiency and effectiveness by shortcutting the probabilistic paths of diffusion models, especially in training one-step diffusion models from scratch (\emph{a.k.a.} shortcut models). However, their theoretical derivation and practical implementation are often closely coupled, which obscures the design space. To address this, we propose a common design framework for representative shortcut models. This framework provides theoretical justification for their validity and disentangles concrete component-level choices, thereby enabling systematic identification of improvements. With our proposed improvements, the resulting one-step model achieves a new state-of-the-art FID50k of 2.85 on ImageNet-256x256 under the classifier-free guidance setting with one step generation, and further reaches FID50k of 2.53 with 2x training steps. Remarkably, the model requires no pre-training, distillation, or curriculum learning. We believe our work lowers the barrier to component-level innovation in shortcut models and facilitates principled exploration of their design space.