AI-native 3D scene generation for simulation-ready robotics and embodied AI.
AnySceneGen builds production-oriented tooling for generating, editing, and validating physically plausible 3D scenes from lightweight inputs. Our goal is to accelerate simulation data creation for robot training, evaluation, and real-world transfer.
Generate sim-ready 3D scenes. Scale Physical AI training.
Powered by spatial intelligence and generative physical simulation, AnySceneGen creates editable, interactive 3D scenes with built-in geometry, materials, semantics, and physical properties—enabling simulation data production at scale for robot training, evaluation, and real-world task generalization.
- Editable — Structured scenes
- Interactive — Sim-ready
- Scalable — Data at scale
- Web Interface (Beta): AnySceneGen is currently available exclusively via our official website (Beta).
- API: Public API access is coming soon.
| Repository | Scope | Maturity | Notes |
|---|---|---|---|
Orbis |
Multi-scale 3D scene dataset covering tabletop, indoor, urban, and natural environments | Dataset / Available | Provides simulation-ready scenes with physical properties, PBR materials, USD/USDC assets, and multimodal edit/customization support |
TabletopGen |
Instance-level interactive 3D tabletop scene generation from text or single image | Research Code / Open Source | Training-free framework for generating diverse, collision-free, simulation-ready tabletop scenes for embodied AI and robotic manipulation |
-
TabletopGen: Instance-Level Interactive 3D Tabletop Scene Generation from Text or Single Image
arXiv -
Orbis: A High-Quality 3D Scene Dataset
Coming Soon
-
IntimeAI — 3D Generative Data Infrastructure
Focus: Building controllable 3D asset and scene generation systems, high-quality simulation-ready datasets, and scalable data infrastructure for embodied AI workflows. -
D-Robotics-AI-Lab — Robotics and Embodied AI Research
Focus: Research on robotic manipulation, 3D scene understanding/generation, visual perception, segmentation, depth estimation, and simulation-oriented embodied AI methods.