The AI painter struggles to generate specific scenes like 'Ice Cola in a teacup,' highlighting the issue of text-image misalignment. Zhao Juntu, a PhD student at Shanghai Jiao Tong University, and his team discovered that even the most advanced AI models have difficulty accurately understanding and realizing complex concepts described in text, such as the difference between a transparent glass cup and a traditional teacup. To address this issue, they proposed a Mixture of Concept Experts (MoCE) method, utilizing large language models to help AI grasp hidden concepts, thereby enabling more precise control over text.