Revisiting Diffusion Model Predictions Through Dimensionality

ArXi:2601.21419v2 Announce Type: replace Recent advances in diffusion and flow matching models have highlighted a shift in the preferred prediction target -- moving from noise ($\varepsilon$) and velocity to direct data (x) prediction -- particularly in high-dimensional settings. However, a formal explanation of why the optimal target depends on the specific properties of the data remains elusive.