VLAForge Official PyTorch implementation of CVPR'26 paper "Unleashing Vision-Language Semantics for Deepfake Video Detection".