With the deepening of the construction of the rule of law and the popularization of artificial intelligence technology, legal intelligence has become a new and important direction in the study of jurisprudence; however, at present, most of the legal intelligence system is presented in the form of text, which lacks visual intuition, and the different groups of people have different degrees of understanding of the law, and the form of text is easy to make part of the masses feel boring, the terminology is obscure and difficult to understand, and the adaptability is poor. In view of the problem, this paper proposes a deep learning-based digital virtual lawyer three-dimensional reconstruction method, the main research through the body type reconstruction, dress reconstruction, facial expression reconstruction and behavioral reconstruction, and other technologies to build three-dimensional digital virtual lawyers. First of all, design the 3D reconstruction model of lawyer’ s position and posture based on convolutional attention CBAM (Convolutional Block Attention Module) and transformer and the lawyer's body shape reconstruction algorithm based on multiscale; then, introduce the clothing matching algorithm and progressive clothing fitting algorithm to carry out the design of digital virtual lawyer’ s progressive clothing modeling; immediately after that, for the reconstruction of digital virtual lawyer’ s facial expression, design the facial surface shape reconstruction, facial surface texture reconstruction, and facial expression synthesis; finally, the imitation of the lawyer’ s behavior is realized by digital virtual lawyer skeleton extraction, designing motion redirection network, and skeleton to video rendering technology. This method has high feature extraction accuracy, overcomes the shortcomings of traditional legal intelligence system which is displayed in the form of text, and helps to solve the problem of visual intuition and extend the adaptability of legal intelligence service.