Shap-e

Shap-e is an open-source generative model from OpenAI for creating 3D assets directly from text prompts or input images. Instead of outputting meshes directly, Shap-e generates neural implicit representations that can be rendered as textured 3D objects from multiple viewpoints. The repository provides pretrained models, inference code, and examples that enable users to generate and visualize 3D content without building models from scratch.

Key capabilities include text-to-3D generation, where users describe an object in natural language, and image-to-3D generation, where a single 2D reference image is used to infer a corresponding 3D shape. Shap-e supports rendering to standard 3D formats and offers utilities for sampling, visualization, and exporting generated assets. The codebase is implemented in Python and integrates with PyTorch, making it suitable for research, experimentation, and extension.