In the domain of video-based action recognition, overcoming catastrophic forgetting while continuously learning new classes remains a major challenge. We propose a Video Class-Incremental Learning (VCIL) framework that addresses this issue by employing a teacher-student knowledge distillation strategy. Our approach leverages both response-based distillation, which aligns the student model’s predictions with the teacher’s softened outputs, and feature-based distillation, which ensures the student retains internal feature representations learned by the teacher. With the UCF101 action recognition dataset and a 3D ResNet backbone, our approach extracts spatiotemporal features to recognize actions in multiple incremental steps. Our model is tested with various settings (10×5, 5×10, 2×25) and has high accuracy for retaining knowledge of past classes and learning new courses. The results show that our approach is efficient in preventing forgetting and maintaining high performance on new tasks.