Yitian Liu, Zhou Zhao, Kun Qi
This study targets the problems of poor semantic consistency and inadequate control over personalized expression in emotion animation generation. Its goal is to develop an automated generation framework that combines multi‑level semantic recognition with style adjustment functions, thereby enhancing the semantic building system and human‑computer interaction paradigms of AIGC technology for controllable emotional animation. Utilizing ChatGPT's NLP capabilities and Runway ML's techniques for generating dynamic images, we construct a hierarchical prompt structure and adopt a three‑dimensional encoding strategy, which allows structured interpretation and coordinated mapping of emotional, behavioral, and stylistic information. The experimental results demonstrated that the proposed method could stably generate animations with coordinated consistency among facial expressions, motion rhythm, and color semantics under various emotion-driven conditions, while exhibiting reasonable temporal progression characteristics. In personalized generation, distinct user profiles produced significant differences in dimensions such as expression amplitude, motion rhythm, and stylistic features, reflecting the model's fine-grained control over multidimensional generation parameters. In the validation of the encoding mechanism, the integration of hierarchical prompts with the multi-level encoding strategy significantly enhanced the naturalness of facial expressions, the logical coherence of movements, the continuity of temporal sequencing, and the precision of micro-action depiction, thereby reinforcing the intensity and realism of emotional presentation. This work develops a framework for generating emotional animations that combines semantic layering with adaptive personalization. It offers methodological backing for the structural optimization and interactive enhancement of AIGC systems within the realm of high‑quality visual content production.