deepfakes/faceswap

FaceSwap is an open-source Python tool that uses deep learning to detect faces in photos and videos, train a model that understands the visual features of two different people's faces, and then swap one face onto the other in new images or footage. The process has three main stages: Extract (automatically detect and crop faces from your source material), Train (feed those cropped face images into a neural network that learns to map one face to another), and Convert (apply the trained model to produce output with the faces swapped). A graphical user interface is also available for users who prefer not to work with the command line. The technology works by using a type of neural network architecture commonly used for image generation tasks. The network learns a compressed representation of each face and then learns to decode that representation as the other face, preserving lighting, expression, and angle. The project includes multiple model types with different quality and performance tradeoffs. You would use FaceSwap for creating visual effects in film or video production, for satire, for educational AI experimentation, or for learning about how deep learning and image synthesis work. The project's maintainers emphasize ethical use strongly and explicitly state it should not be used to create non-consensual content. Running the training effectively requires a modern GPU; CUDA-capable NVIDIA cards are supported, and many AMD cards are supported on Linux via the ROCm framework. The tech stack is Python, with PyTorch or TensorFlow as the deep learning backend, and the project runs on Windows, Linux, and macOS.