Video Input Processing

FaceLift is an innovative tool designed to rapidly generate high-quality 360-degree 3D head models from a single face image. The tool uses a feed-forward approach to generate consistent side and back views of the head from a multi-view latent diffusion model, and then uses the GS-LRM reconstructor to generate a full 3D representation. FaceLift excels on real-world images even when trained only on composite data, supports 4D new-view compositing from video input, and seamlessly integrates with 2D re-animation techniques for 3D facial animation. Key features...