With the rapid advancement of AI, DeepFakes have become increasingly realistic and difficult to detect, posing serious risks such as misinformation, identity theft, and digital manipulation. This project presents a robust DeepFake detection system that combines deep learning and feature engineering techniques to accurately distinguish between real and fake images. By leveraging transfer learning with ResNet-50 along with handcrafted statistical, geometric, and texture-based features, the model captures subtle inconsistencies present in manipulated media. The proposed approach follows a multimodal strategy, integrating both image-based deep learning models and feature-driven machine learning methods, resulting in improved detection performance and generalization.