DiffusionOPD: A Unified Perspective of On-Policy Distillation in Diffusion Models
This work provides a unified and efficient method for multi-task optimization in diffusion-based text-to-image models, addressing cross-task interference and catastrophic forgetting.
DiffusionOPD introduces a multi-task training paradigm for diffusion models that decouples single-task exploration from multi-task integration via online policy distillation, achieving state-of-the-art results on all evaluated benchmarks with better training efficiency and final performance compared to multi-reward RL and cascade RL baselines.
Reinforcement learning has emerged as a powerful tool for improving diffusion-based text-to-image models, but existing methods are largely limited to single-task optimization. Extending RL to multiple tasks is challenging: joint optimization suffers from cross-task interference and imbalance, while cascade RL is cumbersome and prone to catastrophic forgetting. We propose DiffusionOPD, a new multi-task training paradigm for diffusion models based on Online Policy Distillation (OPD). DiffusionOPD first trains task-specific teachers independently, then distills their capabilities into a unified student along the student own rollout trajectories. This decouples single-task exploration from multi-task integration and avoids the optimization burden of solving all tasks jointly from scratch. Theoretically, we lift the OPD framework from discrete tokens to continuous-state Markov processes, deriving a closed-form per-step KL objective that unifies both stochastic SDE and deterministic ODE refinement via mean-matching. We formally and empirically demonstrate that this analytic gradient provides lower variance and better generality compared to conventional PPO-style policy gradients. Extensive experiments show that DiffusionOPD consistently surpasses both multi-reward RL and cascade RL baselines in training efficiency and final performance, while achieving state-of-the-art results on all evaluated benchmarks.