Diffusion fashions obtain high-quality picture technology however are restricted by sluggish iterative sampling. Distillation strategies alleviate this by enabling one- or few-step technology. Circulate matching, initially launched as a definite framework, has since been proven to be theoretically equal to diffusion below Gaussian assumptions, elevating the query of whether or not distillation strategies corresponding to rating distillation switch instantly. We offer a easy derivation — primarily based on Bayes’ rule and conditional expectations — that unifies Gaussian diffusion and stream matching with out counting on ODE/SDE formulations. Constructing on this view, we prolong Rating id Distillation (SiD) to pretrained text-to-image flow-matching fashions, together with SANA, SD3-Medium, SD3.5-Medium/Giant, and FLUX.1-dev, all with DiT backbones. Experiments present that, with solely modest flow-matching- and DiT-specific changes, SiD works out of the field throughout these fashions, in each data-free and data-aided settings, with out requiring instructor finetuning or architectural adjustments. This gives the primary systematic proof that rating distillation applies broadly to text-to-image stream matching fashions, resolving prior issues about stability and soundness and unifying acceleration strategies throughout diffusion- and flow-based turbines.
- † The College of Texas at Austin
