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

arXiv: https://arxiv.org/abs/2609.04282 · HTML: https://mlsanity.com/paper/2609.04282/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Diffusion and flow policies for robotics, one of 14 directions in Diffusion models.
- Builds on 24 of its 42 references that are in the map.
- First posted on arXiv 3 Sep 2026.

## Placed in

- Diffusion models / [Diffusion and flow policies for robotics](https://mlsanity.com/diffusion/diffusion-and-flow-policies-for-robotics/)
- LLM alignment / RLHF / preference optimisation / [Diffusion and multimodal model alignment](https://mlsanity.com/alignment/diffusion-and-multimodal-model-alignment/)

## Builds on (most cited first, 24 of 24)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Denoising Diffusion Implicit Models](https://mlsanity.com/paper/2010.02502/) (2020)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Reflexion: Language Agents with Verbal Reinforcement Learning](https://mlsanity.com/paper/2303.11366/) (2023)
- [CLIPScore: A Reference-free Evaluation Metric for Image Captioning](https://mlsanity.com/paper/2104.08718/) (2021)
- [ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation](https://mlsanity.com/paper/2304.05977/) (2023)
- [Training Diffusion Models with Reinforcement Learning](https://mlsanity.com/paper/2305.13301/) (2023)
- [Flow-GRPO: Training Flow Matching Models via Online RL](https://mlsanity.com/paper/2505.05470/) (2025)
- [DanceGRPO: Unleashing GRPO on Visual Generation](https://mlsanity.com/paper/2505.07818/) (2025)
- [Automatically Correcting Large Language Models: Surveying the landscape of diverse self-correction strategies](https://mlsanity.com/paper/2308.03188/) (2023)
- [HPSv3: Towards Wide-Spectrum Human Preference Score](https://mlsanity.com/paper/2508.03789/) (2025)
- [TempFlow-GRPO: When Timing Matters for GRPO in Flow Models](https://mlsanity.com/paper/2508.04324/) (2025)
- [BranchGRPO: Stable and Efficient GRPO with Structured Branching in Diffusion Models](https://mlsanity.com/paper/2509.06040/) (2025)
- [From Reflection to Perfection: Scaling Inference-Time Optimization for Text-to-Image Diffusion Models via Reflection Tuning](https://mlsanity.com/paper/2504.16080/) (2025)
- [Reflect-DiT: Inference-Time Scaling for Text-to-Image Diffusion Transformers via In-Context Reflection](https://mlsanity.com/paper/2503.12271/) (2025)
- [Zigzag Diffusion Sampling: Diffusion Models Can Self-Improve via Self-Reflection](https://mlsanity.com/paper/2412.10891/) (2024)
- [RealRAG: Retrieval-augmented Realistic Image Generation via Self-reflective Contrastive Learning](https://mlsanity.com/paper/2502.00848/) (2025)
- [Generalizing Alignment Paradigm of Text-to-Image Generation with Preferences through f-divergence Minimization](https://mlsanity.com/paper/2409.09774/) (2024)
- [SymmCompletion: High-Fidelity and High-Consistency Point Cloud Completion with Symmetry Guidance](https://mlsanity.com/paper/2503.18007/) (2025)
- [Weak-to-Strong Diffusion with Reflection](https://mlsanity.com/paper/2502.00473/) (2025)
- [Sculpting Features from Noise: Reward-Guided Hierarchical Diffusion for Task-Optimal Feature Transformation](https://mlsanity.com/paper/2505.15152/) (2025)
- [DetailAnywhere: Fashion Detail Generation via Cross-Modal Feature Alignment Distillation](https://mlsanity.com/paper/2607.02220/) (2026)

## Built on by (most cited first, 1 of 1)

- [KwaiMind Technical Report](https://mlsanity.com/paper/2609.26375/) (2026)

## Abstract (arXiv)

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.