1.x-Distill: Breaking the Diversity, Quality, and Efficiency Barrier in Distribution Matching Distillation

TL;DR

1.x-Distill introduces a fractional-step diffusion distillation framework that enhances quality, diversity, and efficiency in text-to-image models, surpassing prior methods with fewer inference steps.

Contribution

It presents the first fractional-step distillation approach, including a novel analysis of teacher CFG, Stagewise Focused Distillation, and a lightweight cache module, improving performance at extreme steps.

Findings

Achieves better quality and diversity at 1.67 and 1.74 effective NFEs.

Surpasses prior few-step methods in experiments on SD3-Medium and SD3.5-Large.

Provides up to 33x speedup over original sampling methods.

Abstract

Diffusion models produce high-quality text-to-image results, but their iterative denoising is computationally expensive.Distribution Matching Distillation (DMD) emerges as a promising path to few-step distillation, but suffers from diversity collapse and fidelity degradation when reduced to two steps or fewer. We present 1.x-Distill, the first fractional-step distillation framework that breaks the integer-step constraint of prior few-step methods and establishes 1.x-step generation as a practical regime for distilled diffusion models.Specifically, we first analyze the overlooked role of teacher CFG in DMD and introduce a simple yet effective modification to suppress mode collapse. Then, to improve performance under extreme steps, we introduce Stagewise Focused Distillation, a two-stage strategy that learns coarse structure through diversity-preserving distribution matching and refines details with inference-consistent adversarial distillation. Furthermore, we design a lightweight compensation module for Distill--Cache co-Training, which naturally incorporates block-level caching into our distillation pipeline.Experiments on SD3-Medium and SD3.5-Large show that 1.x-Distill surpasses prior few-step methods, achieving better quality and diversity at 1.67 and 1.74 effective NFEs, respectively, with up to 33x speedup over original 28x2 NFE sampling.