point-e

Point-E is an AI system from OpenAI that generates 3D shapes from text descriptions or images. Instead of producing a traditional 3D mesh directly, it creates what is called a point cloud: a collection of thousands of colored dots arranged in three-dimensional space that together represent the shape of an object. A separate model can then convert that point cloud into a mesh, which is the format used by most 3D software.

The typical workflow starts with a text prompt. The system first generates a 2D image of the described object, then uses that image as a guide to produce the 3D point cloud. There is also a pure text-to-3D path that skips the image step, though the README notes this produces lower quality results and handles only simple categories and colors.

The code is released as a Python package and includes Jupyter notebooks (interactive documents that mix code and explanations) to walk through the main use cases: generating a point cloud from an image, generating one from text, and converting a point cloud to a mesh. Evaluation scripts for measuring quality are also included.

The repository accompanies a research paper and is intended mainly for researchers and developers who want to experiment with 3D generation. The README is short and does not cover advanced configuration. Installation is done through pip, the standard Python package installer.