Preprint
Oct 2026
FocusOPD: Structure-Guided On-Policy Distillation for Diffusion Models
FocusOPD is introduced, which augments the base OPD objective with token-wise directional feature alignment restricted to deep layers and early denoising steps, and improves capability transfer in compositional generation, text rendering, and preference-based evaluation.
Xin-Rui Lin, Shu-Min Wang, Su-Qi Song et al.
· 0 citations