XCube
Large-scale 3D generation on sparse voxel hierarchies — up to 1024^3, objects to 100m outdoor scenes.
About
XCube (pronounced X-cubed), a CVPR 2024 highlight from NVIDIA's Toronto lab, generates high-resolution sparse 3D voxel grids with arbitrary per-voxel attributes. A hierarchical voxel latent diffusion model builds grids coarse-to-fine on the VDB data structure, reaching a finest effective resolution of 1024^3 in a single feed-forward pass — no test-time optimization. The demos range from detailed objects to outdoor scenes 100 meters across with 10cm voxels, plus user-guided editing, scene completion from a single scan, and text-to-3D.
Code and pretrained checkpoints are public (checkpoints on Hugging Face), but this is NVIDIA research software: the README routes business inquiries to NVIDIA Research licensing, so treat it as research/non-commercial use. The paper reports clear qualitative and quantitative improvements over prior approaches, but there is no standardized public benchmark table; the numbers are the authors' own.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
No public standardized benchmark table; paper reports qualitative and quantitative gains over prior approaches (author-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: CVPR 2024 highlight from NVIDIA Research; public code and checkpoints (~550 stars); authors include the VDB creator. Strengths: unique sparse-voxel approach, real public checkpoints, scene-scale generation. Docked for: (1) research/non-commercial licensing — not truly open; (2) no independent benchmarks, numbers are author-reported; (3) niche representation (voxels) with thin community tooling. License: NVIDIA research license; commercial use needs NVIDIA's permission.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 5.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 6.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · CVPR 2024 highlight (NVIDIA Toronto, UofT, Vector)
- · Public code + checkpoints on Hugging Face
- · Non-commercial research license; business use via NVIDIA
- · No independent benchmarks