Abstract: The rise of synthetic media, commonly known as deepfakes, has created serious threats to digital authenticity. This paper presents a simple yet effective deepfake detection system for static facial images. It uses transfer learning with the ResNet50 CNN to distinguish between real and manipulated images by detecting subtle differences in facial features. A user-friendly interface was created using Streamlit and deployed through Hugging Face Spaces, allowing for real-time classification. While the system achieves moderate accuracy (about 61%), it shows the potential for accessible and deployable AI tools in digital forensics.

| DOI: 10.17148/IJIREEICE.2025.13713