<p>Artificial intelligence (AI) cloning has been increasingly applied to the creation of deepfake videos, posing significant threats to the legitimacy and authenticity of visual information. This raises concerns about the spread of fake news and undermines people's trust in the digital environment. In this article, we present a novel hybrid deep learning model designed for the efficient detection of deepfakes in videos using the Transfer Learning technique. Recognizing that both spatial and temporal features are critical to detection performance, our model integrates CNN, RNN, Inception and Xception architectures. To evaluate the efficiency of the proposed algorithm, we conducted extensive experiments on three datasets individually: Celeb-DF, FaceForensics++ and DFDC. The results demonstrate the effectiveness of the proposed algorithm, achieving an overall average accuracy around 80%, precision of 79. 87%, recall of 78.07 and an AUC of 0. 80. These findings highlight the effectiveness of the hybrid deep learning approach in detecting deepfake videos and contribute to advancing research in the field of digital media trustworthiness.</p>