From "What" to "How": Constrained Reasoning for Autoregressive Image Generation

TL;DR

CoR-Painter is proposed, a novel framework that pioneers a "How-to-What" Paradigm by introducing Constrained Reasoning to guide the autoregressive generation and introduces a Dual-Objective GRPO strategy that specifically optimizes the textual constrained reasoning and visual projection processes to ensure the coherence and quality of the entire generation pipeline.

Abstract

Autoregressive image generation has seen recent improvements with the introduction of chain-of-thought and reinforcement learning. However, current methods merely specify "What" details to depict by rewriting the input prompt, yet fundamentally fail to reason about "How" to structure the overall image. This inherent limitation gives rise to persistent issues, such as spatial ambiguity directly causing unrealistic object overlaps. To bridge this gap, we propose CoR-Painter, a novel framework that pioneers a "How-to-What" paradigm by introducing Constrained Reasoning to guide the autoregressive generation. Specifically, it first deduces "How to draw" by deriving a set of visual constraints from the input prompt, which explicitly govern spatial relationships, key attributes, and compositional rules. These constraints steer the subsequent generation of a detailed description "What to draw", providing a structurally sound and coherent basis for accurate visual synthesis. Additionally, we introduce a Dual-Objective GRPO strategy that specifically optimizes the textual constrained reasoning and visual projection processes to ensure the coherence and quality of the entire generation pipeline. Extensive experiments on T2I-CompBench, GenEval, and WISE demonstrate that our method achieves state-of-the-art performance, with significant improvements in spatial metrics (e.g., +5.41% on T2I-CompBench).

Figures (14)

• Figure 1: Illustration of the structure and output of (a) CoT-based method, T2I-R1 jiang2025t2i, and (b) our CoR-Painter.
• Figure 2: Overview of CoR-Painter: (a) illustration of the text-to-image generation process, and (b) Dual-Objective GRPO, $R_\text{SA}$, $R_\text{SP}$ and $R_\text{HA}$ represent Semantic Anchoring Reward, Semantic Projection Reward and Holistic Alignment Reward, respectively.
• Figure 3: Overview of the reward computation.
• Figure 4: Comparison of description and generated images before and after incorporating the thought process.
• Figure 5: Qualitative comparison of CoR-Painter against recent methods on typical cases.