Step Back to Move Forward: Reflection-Aware Preference Optimization for Visual Generation

• URL: http://arxiv.org/abs/2609.04282v1
• Date: Thu, 03 Sep 2026 04:37:19 GMT
• Title: Step Back to Move Forward: Reflection-Aware Preference Optimization for Visual Generation
• Abstract summary: Reflection-Aware GRPO is a new RL-based preference alignment framework for diffusion generative models.<n>We introduce Diffusion Reflection, which rectifies intermediate sampling trajectories by inverting the diffusion process with a weak estimator.<n>We also introduce Counterfactual Path Synthesis to implicitly distill these rectified trajectories into the policy, enabling the model to internalize the benefits of search-based exploration.
• Score: 12.58348084349179
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Diffusion models have become the mainstream paradigm for modern visual generation and have substantially advanced multimedia content synthesis, especially in text-to-image and text-to-video tasks. To further align such generative models with human preferences, reinforcement learning (RL) has recently shown strong potential as a post-training strategy. Nevertheless, existing policy gradient-based methods often explore inefficiently, making them vulnerable to local optima that may degrade semantic faithfulness and visual realism. To address these challenges, we present Reflection-Aware GRPO (RA-GRPO), a new RL-based preference alignment framework for diffusion generative models. The core idea is to improve "forward" generation by incorporating "backward" reflection during optimization. We first introduce Diffusion Reflection, which rectifies intermediate sampling trajectories by inverting the diffusion process with a weak estimator, guiding latent states toward higher-probability regions of the true data manifold. Furthermore, we introduce Counterfactual Path Synthesis to implicitly distill these rectified trajectories into the policy, enabling the model to internalize the benefits of search-based exploration without incurring inference-time overhead. Extensive experiments on T2I and T2V models demonstrate that RA-GRPO significantly outperforms existing methods, particularly in mitigating reward hacking and improving generalization. The method remains architecture-agnostic and integrates seamlessly with standard pipelines, suggesting a promising direction for stable preference alignment.

Related papers

• Power Reinforcement Post-Training of Text-to-Image Models with Super-Linear Advantage Shaping [66.25536973294726]
  Post-training methods for text-to-image (T2I) models are prone to reward hacking.<n>Super-Linear Advantage Shaping (SLAS) reshapes the local policy space.<n>SLAS consistently surpasses the DanceGRPO baseline across multiple backbones and benchmarks.
  arXiv  Detail & Related papers  (2026-05-11T17:59:25Z)
• Diffusion-APO: Trajectory-Aware Direct Preference Alignment for Video Diffusion Transformers [12.948398661304184]
  Diffusion-APO is a trajectory-aware algorithm for aligning video diffusion models with human intent.<n>We introduce a unified and modular RLHF framework that integrates online ranking, half-online anchoring, offline refinement, and distillation-aware drift correction.<n>We demonstrate that Diffusion-APO consistently outperforms standard baselines in visual quality and instruction following.
  arXiv  Detail & Related papers  (2026-05-08T09:37:46Z)
• S-GRPO: Unified Post-Training for Large Vision-Language Models [11.901218794806796]
  We propose Supervised Group Relative Policy Optimization (S-GRPO) for adapting Large Vision-Language Models (LVLMs)<n>S-GRPO integrates the guidance of imitation learning into the multi-trajectory exploration of preference optimization.<n>It gracefully bridges the gap between Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL)
  arXiv  Detail & Related papers  (2026-04-17T08:39:07Z)
• Evolution of Optimization Methods: Algorithms, Scenarios, and Evaluations [98.44542103979735]
  First-order gradient descent methods, epitomized by gradient descent (SGD) and Adam, serve as the cornerstone of modern training pipelines.<n>Large-scale model training, stringent privacy requirements, and distributed learning paradigms expose critical limitations in these conventional approaches regarding privacy protection and memory efficiency.<n>We retrospectively analyze the evolutionary trajectory of deep learning optimization algorithms and present a comprehensive empirical evaluation of mainstreams across diverse model architectures and training scenarios.<n>We distill key emerging trends and fundamental design trade-offs, pinpointing promising directions for future research.
  arXiv  Detail & Related papers  (2026-04-14T17:01:36Z)
• UniGRPO: Unified Policy Optimization for Reasoning-Driven Visual Generation [51.41441081823758]
  Unified models capable of interleaved generation have emerged as a promising paradigm.<n>We propose a unified reinforcement learning framework tailored for interleaved generation.<n>Our experiments demonstrate that this unified training recipe significantly enhances image generation quality through reasoning.
  arXiv  Detail & Related papers  (2026-03-24T17:59:17Z)
• Toward Generalizable Deblurring: Leveraging Massive Blur Priors with Linear Attention for Real-World Scenarios [9.82847623835017]
  GLOWDeblur is a Generalizable reaL-wOrld lightWeight Deblur model that combines convolution-based pre-reconstruction & domain alignment module with a lightweight diffusion backbone.<n>We propose Blur Pattern Pretraining (BPP), which acquires blur priors from simulation datasets and transfers them through joint fine-tuning on real data.<n>We further introduce Motion and Semantic Guidance (MoSeG) to strengthen blur priors under severe degradation, and integrate it into GLOWDeblur, a Generalizable reaL-wOrld lightWeight Deblur model that combines convolution-based pre-reconstruction &
  arXiv  Detail & Related papers  (2026-01-10T11:01:31Z)
• Generative Actor Critic [74.04971271003869]
  Generative Actor Critic (GAC) is a novel framework that decouples sequential decision-making by reframing textitpolicy evaluation as learning a generative model of the joint distribution over trajectories and returns.<n>Experiments on Gym-MuJoCo and Maze2D benchmarks demonstrate GAC's strong offline performance and significantly enhanced offline-to-online improvement compared to state-of-the-art methods.
  arXiv  Detail & Related papers  (2025-12-25T06:31:11Z)
• Plug-and-Play Prompt Refinement via Latent Feedback for Diffusion Model Alignment [54.17386822940477]
  We introduce PromptLoop, a plug-and-play reinforcement learning framework that incorporates latent feedback into step-wise prompt refinement.<n>This design achieves a structural analogy to the Diffusion RL approach, while retaining the flexibility and generality of prompt-based alignment.
  arXiv  Detail & Related papers  (2025-10-01T02:18:58Z)
• Rethinking Direct Preference Optimization in Diffusion Models [15.358181258656229]
  We propose a novel approach to enhancing diffusion-based preference optimization.<n>First, we introduce a stable reference model update strategy that relaxes the frozen reference model, encouraging exploration.<n>Second, we present a timestep-aware training strategy that mitigates the reward scale imbalance problem across timesteps.
  arXiv  Detail & Related papers  (2025-05-24T15:14:45Z)
• A Survey of Direct Preference Optimization [103.59317151002693]
  Large Language Models (LLMs) have demonstrated unprecedented generative capabilities.<n>Their alignment with human values remains critical for ensuring helpful and harmless deployments.<n>Direct Preference Optimization (DPO) has recently gained prominence as a streamlined alternative.
  arXiv  Detail & Related papers  (2025-03-12T08:45:15Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.