---
license: mit
tags:
  - image-to-3d
  - trellis2
  - quantized
  - gptq
---

# TRELLIS 2 Quantized (v3 GPTQ) — Standalone Package

Run TRELLIS 2 image-to-3D on **consumer GPUs** (RTX 3060 12GB / 8GB VRAM).

v3 GPTQ quantization: **7.5GB package** instead of 16.2GB FP16. Zero quality loss on standard inputs (LPIPS 0.089 avg vs FP16 baseline).

## Quick Start

```bash
pip install huggingface_hub
huggingface-cli download Lifec-321/trellis2-quantized-v3 --local-dir ./trellis2-quantized
python trellis2_quantized_inference.py --standalone-dir ./trellis2-quantized --image photo.png
```

## Speed Modes

| Flag | GPU Required | Peak VRAM | Speed |
|---|---|---|---|
| `--low-vram` (auto for GPUs with 12GB or less) | 8-12GB | ~4 GB | ~7 min |
| `--fast` (auto for GPUs with 16GB or more) | 16-24GB | ~14 GB | ~6.5 min |

## System Requirements

| Component | Minimum | Recommended |
|---|---|---|
| GPU VRAM | 8 GB | 12+ GB |
| System RAM | 16 GB | 32 GB |
| Disk | 15 GB free | 20 GB free |
| Python | 3.10+ | 3.10 |
| CUDA | 11.8+ | 12.1+ |