Abstract: With the advancement of digital human and virtual interaction technologies, the role of dynamic facial expression generation in personalized content creation has become increasingly prominent. To address the issues of high cost and insufficient controllability in traditional methods, as well as user fatigue often caused by interactive genetic algorithms, this paper proposes a method for dynamic facial expression generation that integrates an implicit keypoint parametric model with a hybrid interactive genetic algorithm. First, based on the concept of implicit keypoint decoupling, facial movements are explicitly represented as low-dimensional parameter vectors with semantic constraints, enabling continuous and controllable modeling of facial expression changes. Second, a structured encoding scheme for dynamic facial expressions is designed. By combining a periodic interaction mechanism with an emotion-recognition-based surrogate model, the method automatically evaluates individual fitness during the evolution process without user involvement, thereby reducing the frequency of user interaction. Furthermore, a comprehensive fitness function is constructed to enhance the plausibility and stability of the generated results. Experimental results demonstrate that the proposed method effectively improves evolutionary efficiency and generation quality while reducing the user interaction burden, exhibiting good robustness and personalized adaptation capabilities in complex emotional expression tasks.