Trellis2 and Pixal3D make empty mesh that cannot be 3D printed
- Dominant language
- Python
- Stars
- 133k
- Forks
- 15.7k
- Avg merge
- 1d 7h
- Merged PRs (30d)
- 158
Description
### Custom Node Testing
- [x] I have tried disabling custom nodes and the issue persists (see [how to disable custom nodes](https://docs.comfy.org/troubleshooting/custom-node-issues#step-1%3A-test-with-all-custom-nodes-disabled) if you need help)
### Expected Behavior
# Hunyuan 3D 2.0 and 2.1
Hunyuan 3D uses "Voxel to Mesh" and "Save 3D Model" make proper STL that can be 3D printed
### Actual Behavior
# Pixal 3D and Trellis 2
Make meshes that have an inner void, this prevents slicers from doing infills to the STLs
The issue is likely not in the mesh->stl conversion
# Fix Attempt
I tried post processing meshes in MeshLab or Blender to no avail. They are quite mangled, hollow and not watertight.
# Root Tracing
This issue is big as it prevents the models from easily being 3D printed. They can still be used as 3D assets.
I tried to root trace the issue, Hunyuan uas a different pipeline, it makes voxels, convert those to meshes and save the mesh, and that works.
Since the issue is shared by Pixal and Trellis, I think the issue is in either the mesh post processing
Another place is the VAE decode stage since they both share the Trellis 2 Shape VAE
### Steps to Reproduce
Run either Pixal 3D or Trellis 2
Save as STL
Try to slice the stl in a slicer like Creality studio
### Debug Logs
```powershell
[INFO] [GetMeshInfo]
Vertices: 3,101,747 (3.10M)
Faces: 6,416,930 (6.42M)
Attributes: none
[INFO] [GetMeshInfo]
Vertices: 404,597 (404.6K)
Faces: 798,508 (798.5K)
Attributes: normals
```
### Other
_No response_
Contributor guide
Research direction
Start by running either the Pixal 3D or Trellis 2 workflow and inspecting the generated mesh before STL conversion. Compare the shared mesh post-processing and Trellis 2 Shape VAE decode stages with Hunyuan 3D's working voxel-to-mesh pipeline. Done means the saved STL is watertight and slicers generate infill correctly.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- computer-graphics, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 48/100