This project implements an AI-based pipeline that automatically generates 3D object files from text or image inputs and renders them in a 3D scene.

By leveraging OpenAI's Shap-E pre-trained model, the system converts textual prompts or input images into 3D latent representations, which are then exported as 3D object files. These generated models are integrated into an OpenGL-based rendering engine for visualization.

To support seamless integration, the rendering engine was modified to handle newly generated object files, including on-the-fly vertex normal computation when normal data is missing. The project was designed to run on a GPU server due to the high computational cost of 3D diffusion models.

This work aims to lower the barrier to 3D content creation by combining AI-driven 3D generation with traditional graphics rendering pipelines.

This project is based on and inspired by OpenAI's Shap-E project, which is released under the MIT License.

Copyright (c) 2023 OpenAI