Traditional patterns are not only an important part of cultural heritage, but also an indispensable part of modern life and art design. The development of diffusion generation technology has brought new opportunities for the preservation, inheritance and innovation of patterns. This paper proposes a top-down structured annotation and generation logic, which establishes the connection between pattern style content and structure from three successively smaller hierarchical perspectives: skeleton structure, layout structure, and element structure, helping the model to accurately capture pattern features and improve learning efficiency. Structured processing also provides support for the diversity and controllability of generation, enabling the model to learn complex skeleton patterns and perform selective visualization generalization based on structural decomposition, providing a reference for the design of such patterns. This paper also established a user-generated pattern platform based on the exploration of patterns, and actually applied it to the design and production of film and television models.