Face presentation attack detection (PAD) is essential for ensuring the security and reliability of face recognition systems by preventing unauthorized access through spoofing attempts. Attackers can exploit various methods, such as printed photos, video replays, paper masks, 3D masks, or makeup, to imitate a legitimate user’s biometric traits. In this paper, we propose an enhanced face PAD solution that leverages the deformable convolutional layer within the MobileNetV2 architecture to improve detection accuracy. By replacing the standard convolution layer with a Deformable ConvNets V2, the proposed model adapts dynamically to spatial distortions, capturing more detailed and robust features for effective face PAD. Extensive experiments on the Replay-Attack, Replay-Mobile, ROSE-Youtu, OULU-NPU, and SiW-Mv2 datasets validate the superiority of the proposed approach. The method achieves a half total error rate (HTER) of 0.0% on both the Replay-Attack and Replay-Mobile datasets, 1.26% on ROSE-Youtu, 4.88% on SiW-Mv2, and an ACER of 0.208% on OULU-NPU, outperforming several existing methods. These results highlight the robustness and effectiveness of our approach in safeguarding face recognition systems against presentation attacks.