Abstract: The primary challenge in developing a domain-specific large model based on a general large model lies in effectively integrating domain-specific knowledge into the general large model. There is a significant amount of tacit knowledge in urban spatial form design, and the public policy nature of urban planning necessitates the efficient communication of tacit knowledge to the public. Leveraging the unique ability of large model to learn and transfer tacit knowledge, this paper proposes a technical framework for integrating urban spatial form design knowledge into general image generation models through domain-specific learning. Tacit and explicit knowledge in spatial form design with design orientation constitutes the core components to guide the domain-specific learning of large models. The integration of urban spatial form design knowledge into a general image generation model is accomplished through the application of large model fine-tuning techniques. Using community public space renewal as a case study, a domain-specific large model incorporating the knowledge of urban spatial form design is constructed. The case shows that the domain-specific large model plays a crucial role in effectively transferring tacit knowledge, enhancing communication efficiency in community renewal, and supporting participatory design processes.