FLAME_PyTorch

Why we included this project

FLAME is a lightweight statistical head model learned from over 33,000 aligned 3D scans, and this repo is the PyTorch implementation that many downstream projects build on. The model combines a linear identity shape space with an articulated neck, jaw, and eyeballs, plus pose-dependent corrective blendshapes and expression blendshapes, so it can represent a face with relatively few parameters. The repository includes the model code and a demo that loads FLAME and visualizes 3D facial landmarks, along with the pose-dependent contour landmarks from RingNet, which makes it usable for fitting the model to scans or images and driving facial animation. It is also the foundation behind well-known systems like DECA, VOCA, and SMPL-X, so it is a practical starting point if you need a compact head model without building one from scratch. One thing to plan for: the model weights are not bundled here, you have to sign up and agree to the FLAME license to download them.