Recent advancements in artificial intelligence, particularly in generative technologies, have significantly redefined the design paradigm for community public cultural spaces, shifting from a traditionally designer-centric model to one that emphasizes multi-stakeholder co-creation. This paper focuses on the design of public cultural spaces at the community scale, proposing a generative dynamic workflow-based co-creation framework that integrates large language models (LLMs) with text-to-image technologies. The framework includes a natural dialogue-based needs-capturing module, a needs analysis module, and a needs expression text-to-image module. This study validates the proposed framework by developing a system prototype for renovating a public space in a student dormitory at Tongji University’s Jiading campus. The results show that the prototype demonstrates good usability and a relatively satisfactory capability in capturing user requirements. These findings indicate that this research helps address key limitations in traditional community design practices, such as limited resident participation, inefficient integration of diverse needs, and slow iteration processes.