GANverse3D Extension


NVIDIA is using AI to power GANverse3D, an Omniverse extension that enables creators to take photos of cars and create virtual replicas with lights, physics models and PBR materials.

Reconstructing virtual assets from just one image is an incredibly hard problem, and we are continually exploring new ideas to increase the resulting asset’s fidelity. Our current AI model is explicitly trained only on sedans, so other types of vehicles will be transformed into sedans. As our research work progresses, expect to see better vehicle diversity as well as improved quality reconstructions.


GANverse3D can be easily installed to an existing Omniverse application through the Extension Manager.

  1. Bring up the Extension Manager by selecting Window->Extensions.

  2. Search for “aitoybox”. Select the aitoybox.ganverse3d extension and click the Install button. (this may take a few minutes, depending on your connection speed)

  3. Enable the extension by clicking on the toggle icon to bring up the GANverse3D main window.

Using GANverse3D

UI Overview

UI Car Settings Legend

Ref #




Add Image

Opens the file picker to select an image of a car


Test Drive

Opens the scene selection dialog. Once a scene is selected, the camera will move to a chase view, physics simulation will be enabled and the car will be respond to inputs



Resets the physics simulation and returns the cars to their original starting point

UI Car Settings Legend

Ref #




Input Image

Selects the 3D car model produced by this input image


Select Visualization

Selects the visualization mode of the car


Toggle Lights

Toggle the car’s lights on/off



Stop the simulation and remove the car from the scene


Change scene

Select and load a new scene

Driving a Car

After selecting a scene through the Test Drive menu, use the keyboard’s arrow keys to drive forward (⬆), backwards (⬇) and turn left (⬅) and right (➡). Hitting the down (⬇) key while driving will cause the car to brake.

Supporting Research

This extension is made possible by the research works the works Image GANs meet Differentiable Rendering for Inverse Graphics and Interpretable 3D Neural Rendering by Yuxuan Zhang, Wenzheng Chen, Huan Ling, Jun Gao, Yinan Zhang, Antonio Torralba, Sanja Fidler, and DatasetGAN: Efficient Labeled Data Factory With Minimal Human Effort by Yuxuan Zhang, Huan Ling, Jun Gao, Kangxue Yin, Jean-Francois Lafleche, Adela Barriuso, Antonio Torralba, Sanja Fidler

Release Notes


  • Add compatibility with Omniverse Create >= 2021.3.8


  • Add compatibility with Omniverse Create >= 2021.3.4

Known Issues

  • Image files stored on a Nucleus server can not be opened directly and must be downloaded to a local drive.