T2I-R1: Reinforcing Image Generation with Collaborative Semantic-level and Token-level CoT

Hongsheng Li (The Chinese University of Hong Kong) · Pheng-Ann Heng (The Chinese University of Hong Kong) · Hao Li (University of Minnesota - Twin Cities) · DONGZHI JIANG (The Chinese University of Hong Kong) · Ziyu Guo (Department of Computer Science and Engineering, The Chinese University of Hong Kong) · Renrui Zhang (The Chinese University of Hong Kong) · ZHUOFAN ZONG (The Chinese University of Hong Kong) · Le Zhuo (Shanghai AI Laboratory) · Shilin Yan (Fudan University)

bi-level reasoningbicot-grpochain-of-thoughtgeneration rewardsjanus-propatch-by-patch generationreasoning-enhanced modelreinforcement learningsemantic-level cotstate-of-the-art modelt2i-compbenchtext-to-image generationtoken-level cotwise benchmark

Recent advancements in large language models have demonstrated how chain-of-thought (CoT) and reinforcement learning (RL) can improve performance. However, applying such reasoning strategies to the visual generation domain remains largely unexplored. In this paper, we present **T2I-R1**, a novel reasoning-enhanced text-to-image generation model, powered by RL with a bi-level CoT reasoning process. Specifically, we identify two levels of CoT that can be utilized to enhance different stages of generation: (1) the semantic-level CoT for high-level planning of the prompt and (2) the token-level CoT for low-level pixel processing during patch-by-patch generation. To better coordinate these two levels of CoT, we introduce **BiCoT-GRPO** with an ensemble of generation rewards, which seamlessly optimizes both generated CoTs within the same training step. By applying our reasoning strategies to the baseline model, Janus-Pro, we achieve superior performance with 13% improvement on T2I-CompBench and 19% improvement on the WISE benchmark, even surpassing the state-of-the-art model FLUX.1. All the training code is in the supplementary material and will be made public.