deepfakes/faceswap

Implements a three-stage pipeline: face extraction using detection models, training generative adversarial networks (GANs) on paired face datasets, and conversion via learned models. Supports GPU acceleration via CUDA and ROCm, with both CLI and GUI interfaces. Modular architecture allows swapping detection and generative models, enabling experimentation with different neural network architectures for improved swap quality.

55,038 stars. Actively maintained with 11 commits in the last 30 days.