ghost

Why we included this project

GHOST is a one-shot face swap model: give it a single source photo and a target image or video, and it transfers that identity across while holding onto the target's expressions and head movement. It comes out of the ai-forever research group with a paper on IEEE Xplore, and the repo includes an inference script, pretrained weights, and a Colab demo, so you can try it before wiring it into anything bigger. The training code and loss controls are also there, useful if you want to fine-tune it on your own portrait set or change how strongly the result favors source identity versus target appearance. One thing worth weighing for production use: the authors explicitly restrict GHOST to ethical, consented scenarios and research, so your pipeline needs to sit comfortably inside that boundary.