With the development of deep learning, the artificial intelligence text generation video model Sora combines the characteristics of Diffusion models, Transformer architecture, and multimodal learning to generate high-quality video content. In this case, we propose an evaluation method for generated videos after cutting based on eigenvectors, which evaluates the AI generated videos after cutting and determines whether the cutted short videos meet the evaluation criteria. This algorithm uses linear algebra knowledge in mathematics to process the eigenvectors of images and calculates the similarity between images using relevant threshold. We calculate the minimum “orthogonal value” of the eigenvectors of the first frame of a generated short video and the last frame of another generated short video, and the accuracy of the evaluation method is 99% when the selected threshold is 0.75. This evaluation method has surpassed neural network algorithms in terms of code simplification and algorithm simplification, achieving good results.