The model is trained on a specific distribution of GAN-generated faces. While it performs strongly on known deepfake patterns, its confidence drops on unseen generators, which reflects the real-world challenge of generalization in deepfake detection. Learn more about GAN-based deepfakes.

This tool provides a probabilistic assessment based on visual patterns learned from known examples of manipulated and AI-generated images. While it can identify common deepfake artifacts, it may not reliably detect all types of manipulated content—especially images generated using unseen or highly advanced techniques. Results should be interpreted with caution and are intended to support human judgment, not replace it.

Fake image examples