“So, does this use FEM?” is often the first question raised whenever someone brings up a new solver or paper. It’s also a regular question we hear when demonstrating AdonisFX. Rather than a simple ‘no, we don’t’, we thought it would be helpful to explain a bit about:
- what FEM is
- how tetrahedrons are distinct from, and independent of, FEM
- character simulation tools today
- Our design approach aka ‘why Adonis doesn’t use FEM’
Let’s jump in.
FEM and Tetrahedrons
Due to the success of character simulation approaches by companies like WetaFX and Ziva Dynamics, FEM (the Finite Element Method) is often conflated with ‘simulation using tetrahedrons’. However, there are many alternative approaches that do not use FEM, but do perform volumetric simulation with tetrahedrons. Similarly, FEM doesn’t require the use of tetrahedrons - they are just one of a variety of element shapes that can be used.
The Finite Element Method is a specific numerical technique for solving partial differential equations by discretising a continuous domain into elements and approximating solutions using basis functions. It originated in structural engineering and scientific computing, and its roots go back to the 1980s in the form most relevant to graphics. FEM is rigorous, well-studied, and designed to prioritise physical accuracy and mathematical convergence.
Tetrahedrons are a geometric representation, a way of discretising a volume into elements to make it possible to perform simulation calculations. FEM is one of many possible methods for solving the equations that govern how those elements behave.
FEM in Media & Entertainment
The most prominent example of FEM in commercial character simulation was Ziva VFX. Co-founded by James Jacobs, who had previously helped develop Weta Digital’s Tissue system, and Jernej Barbic, a leading expert in finite element analysis, Ziva was built from the ground up on an implicit FEM solver.
Tissue won a Scientific and Engineering Academy Award in 2013 and was used on films such as Avatar and The Hobbit. FEM was the common thread connecting both products, and for a period it was treated as the gold standard for anatomically-driven character simulation. Ziva was eventually acquired by Unity in 2022, with sales and support unfortunately ending in April, 2024.
Ziva and Weta’s successes likely led to the perception that all simulations use FEM, but the reality is that many commercial and proprietary solutions do not use FEM. We’ll leave a discussion of proprietary approaches for another day, but we will briefly cover Houdini and AdonisFX.
SideFX Houdini
Perhaps surprisingly, SideFX Houdini doesn’t use FEM in either the Vellum or Otis solvers for character simulation*.
Houdini’s Vellum, which has become the most widely used production simulation framework for cloth, hair, and soft bodies, is built on XPBD (Extended Position-Based Dynamics). When SideFX introduced Vellum in Houdini 17, they described it as a “fast unified solver for cloth, hair, soft bodies and grains” inspired by Nvidia’s XPBD method. The XPBD paper itself (Macklin, Müller, and Chentanez, 2016) introduced iteration-count-independent stiffness, solving one of the core limitations of earlier position-based approaches and making the method viable for production-quality work.
More recently, SideFX introduced the Otis solver for muscles and tissue in Houdini 21. Otis is built on Vertex Block Descent (VBD), a method published at SIGGRAPH 2024 by Chen, Liu, Yin, and Yuksel. SideFX’s own documentation describes the Otis Solver as “built on top of the Vertex Block Descent (VBD) architecture,” enabling “simplified workflows, more robust collisions, and improved anatomical accuracy.”
*note: Houdini does have FEM-based solvers outside of Vellum and Otis
AdonisFX
AdonisFX uses tetrahedral simulation in specific solvers where volumetric representation genuinely serves the result. It does not use FEM. The solving methods we use were selected based on our experiences in production. In particular, we chose to prioritise speed, controllability, stability under extreme deformation, and the ability to iterate without specialist knowledge.
FEM brings several properties that work against the production priorities we identified:
- slow to solve: it prioritises convergence and physical accuracy over interactive feedback.
- hard to art-direct: the mathematical framework resists the kind of intuitive creative controls that artists need.
- difficult to set up, demanding deep technical expertise.
- hard to maintain and extend: the mathematical coupling makes adding new features expensive.
- requires significant and ongoing engineering investment to sustain
- requires exceptional talent to envision architectural improvements.
There have been many great (Oscar-winning!) characters built using FEM-based simulation tools, so none of this is a criticism of any implementation of FEM. These are just the inherent challenging aspects of the method itself. FEM was designed to solve complex structural engineering and elasticity problems in civil and aeronautical engineering and then repurposed to handle character simulation. The alternative methods seen in production (XPBD, VBD, projective dynamics) came to exist because researchers and engineers recognised that character simulation in entertainment has fundamentally different limitations and requirements, and therefore other approaches could achieve similar production outcomes without the same challenges.
Our design philosophy
AdonisFX is focused on two key design ideals:
- The framework should be capable of making creatures and characters that look real.
- The framework should make it quick and easy for artists to build those creatures and characters.
Our experiences from production have shown us that sometimes an accurate, anatomically correct representation of reality doesn’t achieve the artistic purposes of the shot. There is artistry in making something that looks plausible and believable and is compelling to the viewer. And so we have focused on making the tools and workflows that best serve that artistry.
In the hands of talented artists and technicians, any of these tools can produce great looking characters. FEM is just one well-proven approach, but we hope this post has explained that there are other, viable approaches for character simulation that can achieve a similar outcome. Ultimately, all of us are just trying to make great looking characters and creatures (and hopefully having fun doing it!).
TL;DR - no, we don’t use FEM 