pose-preservation

FaithfulFaces is a new framework for text-to-video generation that preserves facial identity consistency across varying poses and occlusions using pose-shared alignment and Euler angle embeddings.