Point-E: A System for Generating 3D Point Clouds from Complex Prompts

ABSTRACT

While text-to-image generation has witnessed rapid progress, text-to-3D synthesis remains challenging due to the lack of massive 3D datasets and the complexity of 3D representations. We introduce Point-E, an efficient system for generating 3D point clouds from text prompts. Rather than training a single end-to-end model, we break the problem into two steps: a text-to-image diffusion model samples a synthetic view, and an image-to-3D diffusion model generates a 3D point cloud conditioned on that view. Point-E generates 3D models in just 1-2 seconds on a single GPU, orders of magnitude faster than existing state-of-the-art methods like DreamFusion, while maintaining high semantic alignment with the prompt.