A Large Language Model-Based System for Semantic Understanding and Automated Scene Generation in Animation Scripts

The development of animation content creation has traditionally relied heavily on manual interpretation of scripts and the labor-intensive design of corresponding scenes. In this paper, we propose a novel system that leverages large language models (LLMs) to achieve semantic understanding of animation scripts and automatic generation of visual scenes and shot sequences. The system integrates natural language understanding, scene graph construction, and multimodal rendering components to translate textual content into temporally and semantically coherent visual representations. Our experimental results demonstrate that the proposed method significantly enhances efficiency and consistency in animation pre-production, reducing human effort and time while preserving creative intent. The study offers new possibilities for intelligent animation workflows and human-AI collaborative storytelling.