---
title: TRELLIS Image to 3D
emoji: 🔮
colorFrom: gray
colorTo: pink
sdk: gradio
sdk_version: 6.1.0
app_file: app.py
pinned: true
license: mit
short_description: Convert image to high-quality 3D model via microsoft/TRELLIS
hardware: zero-gpu-t4
---

# TRELLIS Image to 3D

Convert 2D images into high-quality 3D models using [TRELLIS](https://trellis3d.github.io/), Microsoft's scalable and versatile 3D generation model.

## 🚀 Features

- **Single Image to 3D**: Generate 3D models from a single input image
- **Multi-Image Support**: Use multiple views of an object for better reconstruction (experimental)
- **Multiple Output Formats**:
  - GLB files for use in 3D applications and game engines
  - Gaussian Splatting (.ply) files for advanced rendering
- **Interactive 3D Viewer**: Preview your generated models directly in the browser
- **Automatic Background Removal**: Uses alpha channel or automatic background removal
- **Configurable Generation**: Adjust sampling steps and guidance strength for fine-tuned results

## 📖 How to Use

1. **Upload an Image**: Click on the image input area and select an image, or choose from the example images below
2. **Configure Settings** (optional): Expand "Generation Settings" to adjust:
   - Seed for reproducibility
   - Sparse Structure Generation parameters (Stage 1)
   - Structured Latent Generation parameters (Stage 2)
3. **Generate**: Click "Generate & Extract GLB" to create your 3D model
4. **Download**: Once generation is complete, download the GLB file or extract Gaussian splatting data

## 💡 Tips for Best Results

- **Image Quality**: Use clear, well-lit images with good contrast
- **Alpha Channel**: Images with transparent backgrounds (alpha channel) work best
- **Object Focus**: Ensure the main object is clearly visible and centered
- **Background**: The tool automatically removes backgrounds if no alpha channel is present

## 🔧 Technical Details

- **Model**: [microsoft/TRELLIS-image-large](https://huggingface.co/microsoft/TRELLIS-image-large)
- **Hardware**: ZeroGPU (T4) - GPU resources are allocated on-demand
- **Processing Time**: Typically 2-5 minutes depending on image complexity and GPU availability

## 📚 Resources

- [TRELLIS Project Page](https://trellis3d.github.io/)
- [Paper](https://huggingface.co/papers/2412.01506)
- [Model Card](https://huggingface.co/microsoft/TRELLIS-image-large)

## 📝 Output Formats

- **GLB**: Universal 3D format compatible with most 3D software, game engines, and web viewers
- **Gaussian Splatting (.ply)**: Advanced point-based representation for high-quality rendering

## ⚠️ Notes

- Multi-image mode is experimental and may not produce optimal results for all image sets
- Gaussian splatting files can be large (~50MB) and may take time to download
- Processing requires GPU resources - you may need to wait if all GPUs are in use

---

Built with [Gradio](https://gradio.app/) and powered by [Hugging Face Spaces](https://huggingface.co/spaces)