FaceLift: An Innovative Tool for Generating High-Fidelity 3D Head Models from Single Face Images

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:

• Single Image 3D Reconstruction: Generate high-fidelity 3D head models from a single face image, accurately capturing facial and hair details.
• Multi-view generation: Utilizing a multi-view potential diffusion model, side and back views of the head are generated from a single facial input, ensuring multi-view consistency.
• GS-LRM Reconstruction: The GS-LRM reconstructor converts the generated multi-view image into a 3D Gaussian representation, enabling the synthesis of a new view of the whole head.
• Video Input Processing: Supports video input and processes each frame to generate 3D Gaussian sequences for 4D new-view compositing.
• 3D facial animation: Can be combined with 2D facial animation methods such as LivePortrait to achieve 3D facial animation effects.

Usage:

1. Visit the FaceLift website: Go to the official FaceLift website for details and instructions.
2. Prepare to enter: Provide a face image or video as input material.
3. Run FaceLift: Follow the guidelines provided on the website to run the FaceLift tool to generate a 3D head model or a 4D new view composite.
4. Preview and Export: View the results and export 3D models or animations as needed.

FaceLift's innovative approach enables users to quickly generate high-quality 3D head models from a single face image or video, opening up new possibilities for 3D reconstruction and animation.