In the era of rapid technological advancement, the emergence of Deepfake technology has transformed our interaction with digital content. Deepfakes are sophisticated synthetic media created using deep learning techniques that alter or replace visual and audio elements in images, videos, and audio recordings. While Deepfakes offer potential benefits in entertainment and training, they also raise ethical, social, and security concerns. Many people have been victims of deepfake tools that are widely available online. Such tools can cheat image recognition algorithms. Therefore, an assessment of the dangers of these tools is necessary so that researchers can focus on novel strategies for anti-deepfake. This study evaluates the ability of four famous deepfake tools, namely Deepfakes, Face2Face, Face Swap, and Neural Textures, to cheat deep learning architectures in fake/real face recognition in images. Experimental results show that the Neural Textures tool is the most sophisticated in creating fake faces, which is the most challenging for the considered fake image detection algorithms. In addition, we propose an architecture that can obtain better performance in fake/real face detection with fewer parameters.