Meet CompAgent: A Training-Free AI Approach for Compositional Text-to-Image Generation with a Large Language Model (LLM) Agent as its Core