D-OPSD: On-Policy Self-Distillation for Continuously Tuning Step-Distilled Diffusion Models

Forum topic · 小凯 · 2026-05-08

Summary

D-OPSD is a novel training paradigm for step-distilled diffusion models, addressing the challenge that standard supervised fine-tuning degrades the few-step inference capability of efficient image generation models such as Z-Image-Turbo and FLUX.2-klein. The authors observe that modern diffusion models using LLM/VLM encoders can inherit their encoders' in-context capabilities, which allows fine-tuning to be reframed as an on-policy self-distillation process. During training, the same model acts as both teacher and student under different contexts: the student is conditioned only on text features, while the teacher is conditioned on multimodal features combining the text prompt and the target image. The training objective minimizes the divergence between the two predicted distributions computed over the student's own roll-outs. By optimizing along the model's own trajectory and under its own supervision, D-OPSD enables learning of new concepts, styles, and other adaptations without sacrificing the original few-step generation capacity. Paper: arXiv 2605.05204.

Overview

Field: Computer Vision Authors: Dengyang Jiang, Xin Jin, Dongyang Liu, Zanyi Wang, Mingzhe Zheng, Ruoyi Du, Xiangpeng Yang, Qilong Wu, Zhen Li, Peng Gao, Harry Yang, Steven Hoi Published: 2026-05-06 arXiv: 2605.05204

Abstract

The landscape of high-performance image generation models is currently shifting from the inefficient multi-step ones to the efficient few-step counterparts (e.g., Z-Image-Turbo and FLUX.2-klein). However, these models present significant challenges for directly continuous supervised fine-tuning. For example, applying the commonly used fine-tuning technique would compromise their inherent few-step inference capability.

To address this, the authors propose D-OPSD, a novel training paradigm for step-distilled diffusion models that enables on-policy learning during supervised fine-tuning.

Key Idea

The paper first finds that modern diffusion models where the LLM/VLM serves as the encoder can inherit its encoder's in-context capabilities. This enables framing the training as an on-policy self-distillation process:

During training, the model acts as both the teacher and the student with different contexts.
The student is conditioned only on the text feature.
The teacher is conditioned on the multimodal feature of both the text prompt and the target image.
Training minimizes the two predicted distributions over the student's own roll-outs.

Result

By optimizing on the model's own trajectory and under its own supervision, D-OPSD enables the model to learn new concepts, styles, etc., without sacrificing the original few-step capacity.

---

*Auto-collected on 2026-05-08*