OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Xcube

Xcube is a research model and code repository for large-scale 3D generative modeling with hierarchical sparse voxel grids and latent diffusion. It generates objects and outdoor scenes, with documented applications including unconditional generation, text-to-3D, user-guided editing and scene completion from a single scan.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Xcube — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Xcube when researching or prototyping 3D object and scene generation with hierarchical sparse voxels, especially when its provided datasets and inference examples match your work.

Who it is for

  • Computer-graphics and computer-vision researchers exploring 3D generative models.
  • Developers seeking research code and inference examples for sparse-voxel object or scene generation.

Documented uses

  • Generate ShapeNet chair, car and plane samples from pretrained checkpoints.
  • Run unconditional Waymo scene sampling and visualize scene results.
  • Explore Objaverse text-to-3D inference and visualize generated objects.
  • Study the repository’s staged autoencoder and latent-diffusion training workflows.
Before you choose

Requirements & limits

  • The documented setup supports Linux and uses Conda.
  • The README says fVDB requires a GPU later than Ampere.
  • Inference requires pretrained checkpoints placed in the checkpoints directory.
  • ShapeNet training data is linked; Waymo training data is listed as coming soon.
  • The repository notes that its released implementation differs from the paper: the refinement network is omitted and mesh extraction is post-processing.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up Xcube for inference

Prepare the documented Linux environment and obtain pretrained checkpoints for the repository’s inference examples.

Before you start

  • Linux
  • Conda
  • A GPU later than Ampere for fVDB
  • The Xcube repository and pretrained checkpoints from Hugging Face

Steps

  1. Clone the Xcube repository and create and activate its Conda environment using environment.yml.

  2. Install the documented fVDB revision and the mesh-extraction extension as described in the README.

  3. Download pretrained checkpoints and place them under the checkpoints directory.

  4. Choose a documented inference example and use the repository’s object or scene visualization script for its results.

Expected outcome

A prepared Xcube environment with checkpoints in place for the documented inference workflows.

Guide limitations

  • The README’s automatic checkpoint-download script is marked temporarily unavailable.
  • The Waymo single-scan conditioning example is marked as coming soon.
  • The released code omits the paper’s refinement network and performs mesh extraction as post-processing.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
18 October 2026
Original website title
nv-tlabs/XCube · Official homepage title · retained in its source language
How we review sourcesSuggest a correction