少步出图不再牺牲多样性
扩散模型出图要跑几十步,把步数压到 4 步的老办法有个代价:学生模型只盯着老师最主流的几种结果,同一提示词反复生成,画面越来越像。这篇换了个监督对象——不去对齐「最终输出分布」,而是让学生的采样结果去对齐老师在每个噪声档位上的完整后验均值,并且用一组指数倾斜的均值约束把整个条件分布钉住;数学上他们证明,只要在一小片邻域里对齐这些倾斜均值,老师的完整后验就被唯一确定,不需要再额外训练一个打分网络。落到模型上,SDXL 和 FLUX.2 [klein] 4B 上 4 步方法的 FID 都是它最好,提示词对齐和画质也没掉队。做批量出图、需要同一提示词下多张不重样候选的人,这是能直接换掉现有 4 步蒸馏方案的那种。
📄 原文摘要(英文)
Diffusion distillation accelerates image and video generation by compressing a pretrained teacher's iterative sampling process into a few network evaluations. Distribution-based methods achieve this by matching the student's output distribution to the teacher's. This matching is often driven by reverse-KL objectives, whose mode-seeking bias favors dominant teacher modes at the expense of diversity across samples. We introduce Posterior Matching Distillation (PMD), a new member of the distribution-based distillation family that combines mode-covering posterior sampling with any-step generation. At zero population loss, PMD recovers the teacher's full posterior of clean images at each trained noise level. Its stochastic student supports any-step sampling budgets within the trained timestep grid without further training. To learn these posteriors, PMD matches exponentially reweighted averages of student samples conditioned on a fixed noisy state to corresponding teacher posterior means. Under Gaussian corruption, exponentially tilted posterior mean is a single teacher evaluation at a shifted input. We prove that, matching tilted means over an open neighborhood uniquely identifies the full posterior. PMD thus supervises the entire conditional distribution through a family of mean constraints, without any auxiliary student-score network. Experiments on SDXL and FLUX.2 [klein] 4B demonstrate that PMD achieves the best FID among 4-step methods on both models, while maintaining competitive or leading performance across prompt-alignment and perceptual-quality metrics.