Sample by Step, Optimize by Chunk: Chunk-Level GRPO for Text-to-Image Generation

Yifu Luo, Haoyuan Sun, Xinhao Hu, Penghui Du, Keyu Fan, Bo Li, SiNan Du, Xu Wan, Zhiyu Chan, Bo Xia, Tiantian Zhang, Yongzhe Chang, Kai Wu, Kun Gai, Xueqian Wang

Keywords: GRPO, text-to-image generation, reinforcement learning, flow matching

Abstract: Group Relative Policy Optimization (GRPO) has shown strong potential for flow-matching-based text-to-image (T2I) generation, but it suffers from two key limitations: inaccurate advantage assignment, and the neglect of temporal dynamics of generation. In this work, we argue that shifting the optimization paradigm from the step level to the chunk level can effectively alleviate these issues. Building on this idea, we propose Chunk-GRPO, the first chunk-level GRPO-based approach for T2I generation. The central insight is to group consecutive steps into coherent 'chunk’s that capture the intrinsic temporal dynamics of flow matching, and to optimize policies at the chunk level. In addition, we introduce an optional weighted sampling strategy to further enhance performance. Extensive experiments show that Chunk-GRPO achieves superior results in both preference alignment and image quality, highlighting the promise of chunk-level optimization for GRPO-based methods.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 9026