Efficient hybrid training speeds up image generation models
Efficient Hybrid Distillation: Synergizing Score and Adversarial Objectives for One-Step Diffusion
Computer Vision and Pattern Recognition
Summary
Generating high-quality images with artificial intelligence can be slow, so the authors looked for ways to make it faster. They found that combining two training methods, one good at understanding the big picture and one good at fine details, works best. They designed a system that smartly switches between these methods depending on the situation. Using this approach, their model quickly creates high-quality images in just one step.
What this means in practice
- •For ai developers: Build faster image generation tools that produce high-quality images in a single step by using adaptive hybrid training techniques.
- •For graphics software engineers: Integrate efficient one-step image synthesis methods into creative software to speed up rendering without sacrificing detail.
Authors
Fei Peng, Junqiang Wu, Haoxian Tan, Jun Zhou, Jie Hu, Xiaoming Wei, Jie Guo, Xiu Li
Abstract
Numerous distillation methods have been developed to accelerate diffusion models. Recent research indicates that hybrid strategies---blending different paradigms---yield superior results compared to single-method approaches. However, current works predominantly rely on a simple combination of objectives, leaving the question of how to effectively synergize them largely underexplored. In this paper, we specifically investigate the synergy between Score Distillation and Adversarial Distillation. We reveal that their operating zones are distinct and complementary: Score Distillation is effective in high-noise regimes for capturing the global distribution yet lacks fine-grained supervision in low-noise regions. Conversely, Adversarial Distillation is effective in low-noise regions but is prone to instability and suffers from discriminator failure when applied to high-noise intervals. Leveraging this insight, we propose Efficient Adversarial Score Distillation (EASD), a framework that synergizes these objectives via a regime-adaptive sampling strategy, directing each objective to focus on its effective interval. We further apply this paradigm to DiT-based Flow Matching models. Empirically, our method achieves a promising one-step FID of 1.39 on ImageNet 256$\times$256 using the SiT-XL/2+REPA model. This work provides a concrete guideline for maximizing the potential of hybrid distillation. Our code and model are available at https://github.com/pf0607/EASD .