Advancements in the fields of computer vision and deep learning have enabled the creation of highly realistic images, especially in generating human faces with an unprecedented level of realism. However, the misuse of these capabilities, such as in creating malicious content, has made image manipulation one of the most significant challenges in our daily lives. Therefore, it has become essential to develop innovative methodologies to distinguish between genuine and computer-generated multimedia, which continuously improves in terms of quality and realism. As a result, an effective model has been developed using deep learning techniques, relying on the deep neural network known as Meso Net and the K-nearest neighbors algorithm. This model, referred to as Meso_KNN, is presented in this paper. What distinguishes this model is its focus on important features in facial images that represent vital characteristics of facial manipulation. Additionally, it harnesses the capabilities of K-nearest neighbors for classification, achieving outstanding efficiency in detecting various types of facial manipulation. The model has been tested on a diverse set of facial images collected in the HFF dataset and has achieved an accuracy rate of up to 100 %. It stands as one of the current leading results in this field.