A quick start ComfyUI implementation of TRELLIS.
-
Single and multi-view image processing
-
Optimized for 8GB VRAM
-
Fast inference (~10s)
-
Pre-compiled wheels for quick installation
| Node | Description |
|---|---|
| Load Trellis Model | Downloads and loads the TRELLIS model for inference |
| Trellis Inference | Converts images into 3D Gaussians with dynamic support for single and multiview input with integrated rendering options |
| Save GLB File | Exports 3D models to GLB format ("fast"/"opt" modes) |
| Remove Background (Square) | Uitlity - Preprocesses images with background removal, standardizes the inputs |
Click to see the videos
singleView.mp4
MultiView_VS_SingleView.mp4
| Framework | ComfyUI |
|---|---|
| GPU | 8GB+ VRAM |
| CUDA | 12.4 |
| Python | 3.10 |
| OS | Windows 11 |
Simply clone this repository into your ComfyUI's custom_nodes folder and launch ComfyUI:
cd custom_nodes
git clone https://github.com/Scaryplasmon/ComfTrellis.git
cd ..
python main.py
#or comfy launchDrag and drop these workflows directly into ComfyUI
| Low Quality | High Quality |
|---|---|
 |
 |
| LowQuality_SingleView | HighQuality_SingleView |
| Low Quality | High Quality |
|---|---|
 |
 |
| LowQuality_MultiView | HighQuality_MultiView |
The implementation is optimized for both efficiency and quality:
- 🔄 Automatic wheel installation
- 💨 25-second processing time at maximum settings thanks to early stopping.
- 💾 8GB VRAM optimization
Coming soon
This is a ComfyUI implementation of TRELLIS by Microsoft Research.
Please refer to their repository for the original implementation and research paper.