In the ever-advancing realm of artificial intelligence, the increasing prevalence of deepfake technology raises significant concerns about the generation of highly convincing yet misleading images. This study is dedicated to the detection of deepfakes within static images, employing two distinct methodologies, namely convolutional neural networks (CNN) and CNN with Gabor filters. The first approach harnesses the robust capabilities of CNNs, trained to identify inconsistencies in facial features, compression rates, and other subtle cues indicative of deepfake manipulation, and the model undergoes fine-tuning through transfer learning on the “Celeb-DF: A New Dataset for DeepFake Forensics.” This targeted approach aims to address learning discrepancies introduced during the creation of deepfakes, contributing to the ongoing efforts in deepfake detection. The second approach introduces a novel perspective by integrating Gabor filters into the CNN architecture. The research presents a unified Gabor function capable of generating linear, elliptical, and circular filters. These adaptive Gabor filters seek to address challenges related to the receptive field-model size dilemma in CNNs, providing a versatile solution for diverse data and applications in the realm of deepfake recognition. Both methodologies are meticulously evaluated on benchmark datasets, showcasing their efficacy in recognizing deepfakes. The CNN approach excels in identifying manipulated images, while the CNN with Gabor filter approach adds a nuanced layer of analysis, capturing fine texture details and expanding the model's capabilities. This research offers a comparative analysis of two distinct methodologies, shedding light on their strengths, weaknesses, and potential synergies in the fight against deceptive media manipulation.