Point-E

Point-E is OpenAI’s open-source system for generating 3D objects from text prompts. Given a description like “a traffic cone,” it produces a colored 3D point cloud in a minute or two on a single GPU — dramatically faster than earlier text-to-3D methods. The repository has earned about 6,900 stars on GitHub, is MIT licensed, and was released alongside the Point-E research paper.

Overview

Point-E works in two stages. First, a fine-tuned GLIDE diffusion model generates a single synthetic view of the requested object as an ordinary image. Second, a diffusion model conditioned on that image generates a point cloud of a few thousand RGB points in 3D space. A point cloud is not yet a mesh, so the repo also includes a regression model that converts point clouds into signed distance fields, from which standard tools can extract a textured mesh for use in 3D software or games. OpenAI released the model weights, the synthetic-view and point-cloud training code paths, and example notebooks. While newer text-to-3D systems produce finer detail, Point-E remains a fast, simple and influential baseline — and a practical way to prototype 3D assets.

Key Features

• Text-to-3D generation producing colored point clouds in minutes
• Two-stage pipeline: GLIDE image synthesis plus image-conditioned point diffusion
• Included point-cloud-to-mesh conversion via signed distance fields
• Released model weights and example notebooks from OpenAI
• Runs on a single GPU; far faster than optimization-based text-to-3D methods
• MIT licensed code for research and experimentation

💳 Pricing

Free. Point-E is open source under the MIT license, including the released model weights; there is no paid tier.

Who Is It For

3D artists, game developers and researchers who want a quick open baseline for turning text prompts into draft 3D assets, or who study diffusion models for 3D generation.