DraCo introduces a novel interleaved reasoning paradigm, Draft-as-CoT, for text-to-image generation that leverages both textual and visual content. This approach addresses limitations of existing methods by generating low-resolution draft images for visual planning and verification, significantly improving the generation of rare attribute combinations and overall image quality.

This paper investigates the prevalence and mechanisms of social bias in Large Vision-Language Model (LVLM)-based text-to-image (T2I) models, revealing that system prompts are a key driver. It introduces FairPro, a training-free meta-prompting framework that enables LVLMs to dynamically generate fairness-aware system prompts, significantly reducing bias while maintaining image quality.