This paper presents a novel approach for detecting fake videos by analyzing style latent vectors and their temporal changes in generated videos. The authors discovered that temporally stable deepfake videos exhibit distinct temporal changes in style latent vectors, which are not present in real videos. Their framework utilizes the StyleGRU module, trained using contrastive learning, to represent the dynamic properties of style latent vectors. Additionally, they introduce a Style Attention Module (SAM) that integrates StyleGRU-generated features with content-based features to detect visual and temporal artifacts. The approach is evaluated across various benchmark scenarios, demonstrating superior performance in cross-dataset and cross-manipulation settings. The study also validates the importance of using temporal changes in style latent vectors to enhance the generality of deepfake video detection. The paper includes detailed methodology, experiments, and ablation studies to support the effectiveness of the proposed approach.This paper presents a novel approach for detecting fake videos by analyzing style latent vectors and their temporal changes in generated videos. The authors discovered that temporally stable deepfake videos exhibit distinct temporal changes in style latent vectors, which are not present in real videos. Their framework utilizes the StyleGRU module, trained using contrastive learning, to represent the dynamic properties of style latent vectors. Additionally, they introduce a Style Attention Module (SAM) that integrates StyleGRU-generated features with content-based features to detect visual and temporal artifacts. The approach is evaluated across various benchmark scenarios, demonstrating superior performance in cross-dataset and cross-manipulation settings. The study also validates the importance of using temporal changes in style latent vectors to enhance the generality of deepfake video detection. The paper includes detailed methodology, experiments, and ablation studies to support the effectiveness of the proposed approach.