Complete Deployment Guide & Troubleshooting for TRELLIS v2

Complete Deployment Guide & Troubleshooting for TRELLIS v2 (RTX 4080S / Ubuntu 22.04)

Tested on: Ubuntu 22.04 + RTX 4080 SUPER 32GB + CUDA 12.4 + Python 3.11 + PyTorch 2.6.0

Date: 2026-08-03

TRELLIS version: v2 (latest main branch)

Hi! I just successfully deployed TRELLIS v2 and got example.py running end-to-end. Here's a comprehensive guide covering the full deployment process and all the pitfalls I encountered. Hope this helps others!

✅ Final Result

example.py completed successfully, generating:

• sample.glb (2.4MB) — final 3D model
• sample.ply (31MB) — point cloud
• sample_gs.mp4 — Gaussian splatting render (360° turntable)
• sample_rf.mp4 — Radiance field render
• sample_mesh.mp4 — Mesh render

Full pipeline: Image input → DINOv2 features → Sparse structure sampling → SLat sampling → Gaussian/Radiance field/Mesh rendering → Mesh simplification (291,595 vertices) → Texture baking (2,500 steps) → Output

📋 Prerequisites

表格 Item | Requirement -- | -- OS | Ubuntu 22.04 (Linux only) GPU | VRAM ≥ 8GB (12GB+ recommended) CUDA | 12.x Disk | ≥ 20GB free space Network | GitHub & HuggingFace access (mirrors needed in China)

🔗 One-Click Deploy Script

I've also prepared an automated deployment script that handles all 13 steps with idempotent re-runs:

👉 deploy_trellis2.sh (link to be added)

Hope this saves someone hours of debugging! 🎉

First identify where deployment documentation belongs in the repository, then review the issue's example.py workflow and the proposed deploy_trellis2.sh script. Confirm the documented Ubuntu, GPU, CUDA, Python, PyTorch, disk, and network prerequisites, and ensure the completed guide includes reproducible troubleshooting steps and a usable script link.