---
license: mit
pipeline_tag: image-to-3d
---

# LGM Full

# NLP to 3D Model – Custom Pipeline

## Overview

This project showcases an experimental pipeline that bridges **natural language prompts to 3D model generation** using a modified version of a pre-trained multi-view diffusion model.

It is part of a final year project for the *Comprehensive Creative Technologies Project* at UWE Bristol. The primary aim was to explore the potential of AI-assisted 3D content creation using natural language input.

---

## Model Source & Attribution

This project **relies on the pre-trained model from the [LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content Creation](https://huggingface.co/papers/2402.05054)**, developed for the **ML for 3D Course** by researchers at Google Research.

🔗 Original Model: *[https://huggingface.co/spaces/dylanebert/LGM-tiny]*  
📄 Paper: [arXiv:2402.05054](https://arxiv.org/abs/2402.05054)  
🔒 License: MIT

I do not claim authorship of the model architecture or training process. This space serves as a **custom wrapper** for experimentation with **text-to-3D workflows**.

---

## What This Model Does

- Allows input of a natural language description
- Internally maps the input to a representative image or multi-view description
- Generates a 3D model using the LGM pipeline

---

## Limitations

- This is a prototype for academic use only.
- The model’s ability to handle complex or abstract text is limited.
- Performance and quality depend entirely on the base pre-trained model.

---

## Acknowledgements

Thanks to Hugging Face and the authors of LGM for making their models publicly available.

---

## Author

**Gordon CHIN HO AU**  
Final Year BSc Digital Media  
University of the West of England, Bristol