Diffusion-SDPO: Safeguarded Direct Preference Optimization for Diffusion Models
By: Minghao Fu , Guo-Hua Wang , Tianyu Cui and more
Potential Business Impact:
Makes AI art follow your words better.
Text-to-image diffusion models deliver high-quality images, yet aligning them with human preferences remains challenging. We revisit diffusion-based Direct Preference Optimization (DPO) for these models and identify a critical pathology: enlarging the preference margin does not necessarily improve generation quality. In particular, the standard Diffusion-DPO objective can increase the reconstruction error of both winner and loser branches. Consequently, degradation of the less-preferred outputs can become sufficiently severe that the preferred branch is also adversely affected even as the margin grows. To address this, we introduce Diffusion-SDPO, a safeguarded update rule that preserves the winner by adaptively scaling the loser gradient according to its alignment with the winner gradient. A first-order analysis yields a closed-form scaling coefficient that guarantees the error of the preferred output is non-increasing at each optimization step. Our method is simple, model-agnostic, broadly compatible with existing DPO-style alignment frameworks and adds only marginal computational overhead. Across standard text-to-image benchmarks, Diffusion-SDPO delivers consistent gains over preference-learning baselines on automated preference, aesthetic, and prompt alignment metrics. Code is publicly available at https://github.com/AIDC-AI/Diffusion-SDPO.
Similar Papers
Towards Self-Improvement of Diffusion Models via Group Preference Optimization
CV and Pattern Recognition
Makes AI pictures better by learning from groups.
Smoothed Preference Optimization via ReNoise Inversion for Aligning Diffusion Models with Varied Human Preferences
CV and Pattern Recognition
Makes AI art better by learning what people like.
Beyond Reward Margin: Rethinking and Resolving Likelihood Displacement in Diffusion Models via Video Generation
CV and Pattern Recognition
Makes AI videos better by learning what people like.