This paper introduces the IRIS Benchmark, a novel framework designed to synchronously evaluate fairness in both understanding and generation tasks within Unified Multimodal Large Language Models (UMLLMs). It aims to resolve the "Tower of Babel" dilemma of fragmented fairness metrics by offering a multi-dimensional, trade-off analysis approach and uncovering systemic biases in leading models.

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.