How AI describes VeridisQuo

VeridisQuo is a neural network model designed to detect deepfake manipulations in face images. It uses a dual-stream architecture that combines spatial features from EfficientNet-B4 with frequency-domain features extracted via DCT and FFT to identify artifacts that deepfakes often leave in the frequency domain. The model, implemented in PyTorch (~25M parameters, ~101 MB), is trained on FaceForensics++ (C23) for binary FAKE/REAL classification and is available on Hugging Face with documentation and usage details.