Add `pytetwild` (fTetWild) tetrahedralization node for FEM / volumetric workflows

[antoinebou12]

Hi! First off—GeometryPack is already super useful for getting “clean” surface meshes inside ComfyUI (repair/remesh/unwrap/preview). I’m trying to push the pipeline one step further into simulation, and I’m missing one key building block:

Please add a tetrahedralization (“tet mesher”) node based on pytetwild (Python wrapper of fTetWild / TetWild).

• Node name idea: Tetrahedralize (TetWild / fTetWild)
• Input: triangle surface mesh (watertight preferred; optionally auto-fix using existing repair nodes)
• Output: tetrahedral volume mesh (vertices + tet connectivity), plus optional surface extraction

Why this matters (FEM / physics sim)

Most FEM solvers (soft-body, stress/strain, heat, modal analysis, etc.) need a volumetric tetra mesh, not just a surface triangle mesh. Having tetrahedralization inside GeometryPack would make ComfyUI a complete mesh-to-sim-ready preprocessing graph:

Example workflow

1. Load mesh
2. Repair / make watertight
3. Remesh / simplify
4. ✅ Tetrahedralize (TetWild)
5. Export to solver format (or visualize quality)

That’s a big win for anyone doing FEM in external tools (Gmsh/FEniCS/SOFA/CalculiX/Abaqus/etc.) or future in-Comfy physics nodes.

Suggested backend: pytetwild

pytetwild is a Python wrapper around fTetWild and ships wheels on PyPI (so users don’t need to compile C++ locally in most cases).

Refs:

• PyPI: https://pypi.org/project/pytetwild/
• fTetWild repo: https://github.com/wildmeshing/fTetWild
• Paper (SIGGRAPH 2020): https://doi.org/10.1145/3386569.3392385

Parameters to expose (matching pytetwild)

These map cleanly to a ComfyUI node UI and are already documented in pytetwild:

• edge_length_fac (default ~0.05 of bbox diagonal)
• edge_length_abs (override)
• optimize (quality vs speed)
• simplify (preprocess surface)
• epsilon (envelope / feature fidelity)
• stop_energy, num_opt_iter
• coarsen
• num_threads
• quiet/loglevel (optional)

Output / interop

For FEM users, exports matter as much as generating tets. Any of these would be awesome:

• Export .msh (Gmsh) or .vtu/.vtk (ParaView) or .xdmf
• At minimum: provide raw arrays (V, T) + a “save” node later

Implementation notes (high level)

• Convert GeometryPack mesh representation → (vertices, faces) numpy arrays
• Call pytetwild.tetrahedralize(vertices, faces, ...)
• Return a new “tet mesh” object type (or a dict payload) that downstream nodes can consume
• Optional: surface extraction node (tet boundary triangles) for preview

Dependency / env note

If you’d rather isolate this dependency: GeometryPack already uses comfy-env patterns for dependency management and isolation, so this could be added either in the main env or as an isolated sub-env if needed.

---

If this is something you’d accept, I can help test on a few nasty meshes (non-manifold, holes, self-intersections) and share expected outputs + solver compatibility notes.
Thanks!