Diffusion-SDPO: Safeguarded Direct Preference Optimization for Diffusion Models
Abstract
Diffusion-SDPO improves text-to-image generation quality by adaptively scaling the loser gradient in preference optimization, ensuring the preferred output's error does not increase.
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.
Community
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- Free Lunch Alignment of Text-to-Image Diffusion Models without Preference Image Pairs (2025)
- Ranking-based Preference Optimization for Diffusion Models from Implicit User Feedback (2025)
- HiGS: History-Guided Sampling for Plug-and-Play Enhancement of Diffusion Models (2025)
- PCPO: Proportionate Credit Policy Optimization for Aligning Image Generation Models (2025)
- MIRA: Towards Mitigating Reward Hacking in Inference-Time Alignment of T2I Diffusion Models (2025)
- CARINOX: Inference-time Scaling with Category-Aware Reward-based Initial Noise Optimization and Exploration (2025)
- IMG: Calibrating Diffusion Models via Implicit Multimodal Guidance (2025)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment:
@librarian-bot
recommend
Models citing this paper 1
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper