SimSwap is an open-source face-swapping framework that can swap an arbitrary face onto images and videos using a single trained model — no per-person training required. Given a source face and a target photo or video, it transfers the source identity while preserving the target’s pose, expression and lighting. The project, from an ACM Multimedia 2020 paper, has more than 5,100 stars on GitHub, with code updates through 2024.
Earlier face-swap systems typically had to be trained on each pair of people, which limited them to celebrities with lots of footage. SimSwap generalised the task: an identity encoder extracts who the source person is, and an ID injection module blends that identity into the target’s features inside a GAN generator, while weak feature matching losses keep the target’s expressions and gaze intact. The result is a framework that performs arbitrary face swapping — any source, any target — on both still images and video frames with one model. The repository includes training and inference code and pretrained checkpoints, and SimSwap became one of the most referenced open implementations in the face-swap literature, widely forked for research and creative experiments. Like all face-swap technology it carries obvious misuse potential, and the authors release it for research; anyone using it should respect consent and local law.
Free. SimSwap is released openly for research purposes; review the repository’s terms before use.
Computer-vision researchers studying face manipulation and generation, and developers prototyping face-swap features who need a proven open implementation.