Unifying Distributional Training for One-Step Visual Generation
MGFlow unifies one-step visual training and beats FD-Loss on ImageNet and four-step FLUX.2.
50
MGFlow unifies one-step visual training and beats FD-Loss on ImageNet and four-step FLUX.2.
Persistence Forcing uses specialized diffusion features to reach FID 1.63 on ImageNet 256.
Today 1.2K readers · 1.4K views · 193 reads · all time 203K views