NeuRiPhy: Neural Baking of Physics-Based Deformations for Facial Rigs
Published in Eurographics/SIGGRAPH Symposium on Computer Animation (SCA), 2025
We introduce a self-supervised neural approach for real-time, physically-principled facial rigs that incorporate soft tissue deformations and contact interactions. Traditional artist-crafted blendshapes lack the capacity to produce realistic tissue deformations, while physics-based models, despite their accuracy, are computationally prohibitive for interactive applications. Our method addresses these limitations by learning a neural map from a set of rig controls to corresponding deformations that minimize the mechanical energy of an anatomically-based face model, including soft tissue, skull, and teeth layers. This self-supervised framework eliminates the need for predefined simulation data and supports unsupervised handling of collisions between rigid and soft bodies, streamlining the computational process. We demonstrate that, for the first time, our approach achieves real-time performance of physics-based face rigs with complex non-linear deformations and contact handling.
Downdload Paper
BibTeX Record:
@article{10.1145/3747866,
author = {Corigliano, Davide and Peter, Daniel and Huber, Niko Benjamin and Thomaszewski, Bernhard and Solenthaler, Barbara},
title = {NeuRiPhy: Neural Baking of Physics-Based Deformations for Facial Rigs},
year = {2025},
issue_date = {August 2025},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {8},
number = {4},
url = {https://doi.org/10.1145/3747866},
doi = {10.1145/3747866},
abstract = {We introduce a self-supervised neural approach for real-time, physically-principled facial rigs that incorporate soft tissue deformations and contact interactions. Traditional artist-crafted blendshapes lack the capacity to produce realistic tissue deformations, while physics-based models, despite their accuracy, are computationally prohibitive for interactive applications. Our method addresses these limitations by learning a neural map from a set of rig controls to corresponding deformations that minimize the mechanical energy of an anatomically-based face model, including soft tissue, skull, and teeth layers. This self-supervised framework eliminates the need for predefined simulation data and supports unsupervised handling of collisions between rigid and soft bodies, streamlining the computational process. We demonstrate that, for the first time, our approach achieves real-time performance of physics-based face rigs with complex non-linear deformations and contact handling.},
journal = {Proc. ACM Comput. Graph. Interact. Tech.},
month = aug,
articleno = {59},
numpages = {18},
keywords = {Neural Simulation, Physics-Based Facial Deformation, Facial Rigs}
}