Under the background of continuous integration of digital art generation and intelligent inheritance of traditional culture, how to maintain the style of meticulous flower-and-bird painting and accurately express national cultural elements in Chinese painting creation has become an important topic of intelligent art research. This paper constructs a deep learning method framework for Chinese painting creation, and designs the system from visual semantic feature representation, multi-modal feature extraction and fusion, style generation constraints to interactive feedback closed loop. The experiment is based on 3600 valid images and 10800 groups of samples after amplification. The results show that the style similarity of the complete model reaches 0.901, the accuracy of cultural element expression reaches 91.8%, the coordination degree of composition reaches 89.6%, and the user satisfaction score reaches 9.1. In addition, the accuracy of cultural expression still maintains 86.7% under 30% disturbance. The research shows that this method can improve the style stability, cultural consistency and creation auxiliary value of digital generation of meticulous flower-and-bird painting, which has positive significance for the intelligent creation of Chinese painting and the digital communication of traditional culture.
