Personalized text-to-image(T2I) generation aims to create customised images with user-specified themes, objects, or styles, conditioned on a given text description. While current state-of-the-art T2I generation models demonstrate impressive image fidelity guided by textual prompts, they still face challenges in fully capturing the semantic richness and contextual diversity implied in natural language. In this paper, we propose a personalized T2I generation framework that uniquely leverages an interaction mechanism between a large language model (LLM) and a personalization model. By embedding identity information within a semantically enhanced diffusion model, our method enables the generation of highly realistic and coherent scenes for specific objects. The resulting images feature both rich contextual backgrounds and accurate subject representations. Experimental results demonstrate that the proposed T2I generative model excels in superior image quality and performance metrics, attributed to the fine-tuning of the LLM on a personalized dataset.