ComfyUI Trellis 2 Setup: Complete Local Installation Guide (2026)

Why Run Trellis 2 Locally with ComfyUI?

Running trellis-2 through ComfyUI gives you powerful advantages:

• Unlimited generations — No API limits or costs
• Custom workflows — Chain with other AI tools
• Privacy — Your images never leave your computer
• Speed — No network latency, direct GPU access
• Control — Fine-tune every parameter

Prerequisites Before You Start

Hardware Requirements

Software Requirements

• Operating System: Linux (Ubuntu 22.04 recommended)
• CUDA Toolkit: 11.8 or 12.2
• Python: 3.10 or 3.11
• ComfyUI: Latest version
• Conda/Mamba: For environment management

Step 1: Install ComfyUI (If Not Already Installed)

# Clone ComfyUI
git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Linux/Mac
# OR: venv\Scripts\activate  # Windows

# Install dependencies
pip install -r requirements.txt

# Test ComfyUI
python main.py

Visit http://127.0.0.1:8188 to verify ComfyUI is running.

Step 2: Install Trellis 2 Custom Nodes

Method A: ComfyUI Manager (Easiest)

1. Install ComfyUI Manager if not installed
2. Open ComfyUI → Manager → Install Custom Nodes
3. Search for "Trellis" or "TRELLIS"
4. Click Install
5. Restart ComfyUI

Method B: Manual Installation

# Navigate to custom nodes directory
cd ComfyUI/custom_nodes

# Clone Trellis nodes
git clone https://github.com/microsoft/TRELLIS.git
cd TRELLIS

# Install dependencies
pip install -r requirements.txt

# Install additional dependencies
. ./setup.sh --basic --xformers --flash-attn --diffoctreerast --spconv --mipgaussian --kaolin --nvdiffrast

Step 3: Download Trellis 2 Models

The models are hosted on Hugging Face. Download these to your ComfyUI models folder:

# Create model directory
mkdir -p ComfyUI/models/trellis

# Download using huggingface-cli
pip install huggingface_hub
huggingface-cli download microsoft/TRELLIS-image-large --local-dir ComfyUI/models/trellis/TRELLIS-image-large

# OR download manually from:
# https://huggingface.co/microsoft/TRELLIS-image-large

Available Models

Step 4: Create Your First Trellis 2 Workflow

Basic Image-to-3D Workflow

1. Load Image Node → Connect to Trellis Input
2. Trellis Image-to-3D Node → Configure settings
3. Save 3D Model Node → Export GLB/PLY

Node Configuration

Trellis Image-to-3D Settings:
├── seed: 42 (or random)
├── sparse_structure_steps: 12
├── sparse_structure_cfg: 7.5
├── slat_steps: 12
├── slat_cfg: 3.0
├── simplify: 0.95
└── texture_size: 1024

Step 5: Optimize for Your GPU

For 16GB VRAM (RTX 4080, A4000)

# Add to ComfyUI launch
--lowvram --preview-method auto

Adjust Trellis settings:

• Reduce texture_size to 512
• Lower resolution to 768
• Enable memory optimization flags

For 24GB+ VRAM (RTX 4090, A100)

Run at full settings:

• texture_size: 2048
• resolution: 1024
• All quality options enabled

Advanced Workflow: Chaining Trellis 2 with Other Nodes

Workflow: Image → Remove Background → Trellis 2 → Export

Load Image
    ↓
Background Removal (RMBG)
    ↓
Trellis 2 Image-to-3D
    ↓
[Optional] Mesh Simplification
    ↓
Save GLB + Preview Video

Workflow: Text → Image (SDXL) → Trellis 2 → 3D

CLIP Text Encode
    ↓
KSampler (SDXL)
    ↓
VAE Decode
    ↓
Trellis 2 Image-to-3D
    ↓
Save 3D Model

Troubleshooting Common Issues

"CUDA out of memory"

Solutions:

1. Close other GPU applications
2. Use --lowvram flag
3. Reduce texture_size and resolution
4. Try CPU offloading for some operations

"Module not found" errors

# Reinstall dependencies
cd ComfyUI/custom_nodes/TRELLIS
pip install -r requirements.txt --force-reinstall

"Flash attention not available"

# Install flash-attn manually
pip install flash-attn --no-build-isolation

# OR use xformers backend
export ATTN_BACKEND=xformers

Slow generation speed

• Ensure CUDA is properly detected: python -c "import torch; print(torch.cuda.is_available())"
• Use SSD storage for models
• Close background applications

Performance Optimization Tips

Speed Optimization

Quality Optimization

FAQ: ComfyUI + Trellis 2

Can I run Trellis 2 on Windows?

Windows support is experimental. Linux is strongly recommended. If you must use Windows, consider WSL2 with Ubuntu.

How much disk space do I need?

Plan for 50-100GB: ~15GB for models, ~30GB for dependencies, plus working space for outputs.

Can I use Trellis 2 with ControlNet?

Yes! Preprocess your image with ControlNet first, then feed the result to Trellis 2 for 3D generation.

Does ComfyUI Trellis support batch processing?

Yes, connect multiple images to process them in sequence. Useful for creating asset libraries.

Can I fine-tune Trellis 2 in ComfyUI?

Training requires separate setup outside ComfyUI. Use the official training scripts from the TRELLIS repository.

Next Steps

Now that you have trellis-2 running locally with ComfyUI:

1. Experiment with different image types and settings
2. Build custom workflows combining multiple AI tools
3. Join the community — Share your workflows on ComfyUI forums
4. Optimize your setup for your specific hardware

Ready to create 3D assets without limits? Your local trellis-2 setup is now complete!

Related Tutorials

• Trellis 2 Complete Tutorial
• Trellis 2 vs Meshy vs Hunyuan: Comparison
• Trellis 2 for Blender Workflow