This version was trained on a collection of face swap images, most of them gone wrong. It works better at higher weights and/or higher CFG scales.
A series of experimental LoRAs trained to alter and warp faces in various ways.