Hybrid Deepfake Detector combining a ResNet-50 CNN and a Vision Transformer (ViT-B/16) to identify AI-generated images. It combines two layers to increase accuracy: a Layer 1 basic EXIF/PNG metadata scanner to catch raw generations, and a Layer 2 PyTorch vision model for deep structural analysis.

Note: The model weights (~423 MB) are securely hosted on Hugging Face. The first time you run predict.py, the script will automatically download the step_APP_READY.pth checkpoint to your local machine. You do not need to download it manually. The GitHub repo is: https://github.com/SantiagoSaldanaS/ai_image_detection