Visual Language Models (VLMs) have transformed the landscape of AI-generated art, opening new possibilities for creative expression and human-AI collaboration. Our research introduces the Identification-Explanation-Implication (IEI) framework, which draws inspiration from artistic practices and cognitive science to analyze and enhance AI's creative capabilities. By studying artist-generated visual mashups—where elements from different sources are deliberately combined to create new meaning—we investigate how VLMs understand and generate creative combinations. Our findings reveal that while AI systems can effectively comprehend creative combinations, often surpassing casual observers, they still fall short of matching expert artists' nuanced understanding. More importantly, we demonstrate that by explicitly incorporating artists' conceptual blending processes into AI systems' generation pipeline, we can significantly improve their creative outputs. This work offers both theoretical insights for artists working with AI and practical guidelines for developing more sophisticated AI art tools, contributing to the ongoing dialogue between traditional artistic practices and emerging AI capabilities.