🎨 FlexAvatar Code and Weights Released
Developers have introduced FlexAvatar—a method for training full 3D head avatars using partial supervision, accepted at CVPR 2026. The project includes open-source code on GitHub and model weights, allowing for the creation of animated 3D portraits based on static images or videos.
🌍 Moving toward partial supervision for creating high-quality 3D avatars reduces the requirements for expensive datasets, simplifying the process of generating digital twins.
👤 It is now possible to create realistic 3D avatars from ordinary photos or videos using open-source software. However, risks remain regarding the unauthorized use of biometric data without the subjects' consent.
Source 1: https://github.com/tobias-kirschstein/flexavatar Source 2: https://tobias-kirschstein.github.io/flexavatar/