Description

GET3D is NVIDIA’s generative model for creating high-quality 3D textured shapes learned from images. It directly synthesizes explicit textured meshes with rich geometric detail and complex topology.

The model generates a 3D SDF and a texture field from latent codes, then uses differentiable surface modeling and rendering to train on 2D image collections. The page highlights applications such as generating diverse 3D assets across categories like cars, chairs, animals, motorbikes, human characters, and buildings.

How we innovate

GET3D is distinctive for generating explicit textured 3D meshes directly from images, using separate latent codes for geometry and texture and training end-to-end with 2D adversarial losses and differentiable rendering.

Use Case / Scenario

1. 3D Asset Generation

Create diverse textured 3D meshes for categories such as cars, chairs, animals, motorbikes, human characters, and buildings.

2. Content Creation for Virtual Worlds

Generate 3D content that can be directly consumed by 3D rendering engines for downstream applications.

3. Text-Guided Shape Exploration

Use text prompts to finetune the 3D generator and produce a range of meaningful shapes from rendered images.

Visit Website