The advancement of urban digital twins depends on the accurate representation of 3D city models, particularly Level of Detail 3 (LoD3) models, which incorporate detailed façade features essential for urban planning applications. However, generating LoD3 models is challenging due to the complexities of semantic segmentation in 3D point cloud data and the high resource demands of traditional methods. This paper presents an automated methodology for upgrading existing Level of Detail 2.2 (LoD2.2) building models to LoD3 using mobile mapping point cloud data and the Grounding DINO model. The approach begins with extracting façade surfaces from LoD2.2 models while maintaining geometric integrity. Point cloud data is then transformed into a 2D image format to facilitate the application of Grounding DINO, which accurately detects and segments façade elements such as windows and doors. The identified features are re-integrated into the 3D model, resulting in an enhanced LoD3 representation. This methodology demonstrates effectiveness and scalability, providing a practical solution for improving urban digital twins with detailed and reliable building models.