In recent years, deep learning technology has become a core driving force in animation production, effectively overcoming the bottlenecks of low efficiency, high cost, and lack of realism in traditional animation production. This paper reviews the application of deep learning in animation production from three perspectives: application, technical advantages and challenges, and future development directions. In terms of application, this paper focuses on two major directions: facial animation and 3D scene generation, and systematically analyzes the algorithm design and practical effects of technologies such as expression parameter regression, voice-driven animation, and convolutional neural network-based relief modeling. In terms of technical advantages, deep learning significantly improves the efficiency and expressiveness of animation production while reducing reliance on professional hardware. However, existing technologies still face challenges such as data sensitivity and insufficient detail reproduction. Future development directions include multimodal fusion, lightweight model deployment, and human-computer collaboration. Multimodal fusion can generate animation scenes by combining text descriptions, improving the efficiency and flexibility of creation. Lightweight model deployment aims to reduce the computational complexity and resource requirements of models. Human-computer collaboration fully utilizes human creativity and AI's technical advantages to achieve higher-quality animation scene generation
X. Dong (Wed,) studied this question.