FactorPortrait: Controllable Portrait Animation via
Disentangled Expression, Pose, and Viewpoint

1Meta Reality Labs 2Technical University of Munich

Studio Dataset

It is a multi-view video dataset captured in a professional studio, similar to Ava-256. It captures diverse facial expressions, head movements, and gaze directions. Each video is captured from a fixed viewpoint. In this dataset, we want to generate static novel view videos with changes in pose and expression.

Input GAGA CAP4D HunyuanPortrait Ours GT
Input GAGA CAP4D HunyuanPortrait Ours GT