Open 3D indoor-scene generation model
SpatialGen
Generates spatially consistent indoor 3D scenes from an image or a text-conditioned layout workflow.
01 / Overview
Indoor geometry and appearance, layout-conditioned generation, multi-view consistency, 3D Gaussian reconstruction.
Generates spatially consistent indoor 3D scenes from an image or a text-conditioned layout workflow.
Check walls, openings, furniture placement, scale, and camera consistency across views.
Keep research outputs and commercial platform features documented separately.
02 / Strengths
Where it stands out
- Focuses on room-scale spatial structure rather than only generated camera motion.
- Open release allows technical inspection and local experimentation.
- Connects world generation with a broader reconstruction, editing, and spatial-understanding platform.
03 / Boundaries
What not to overclaim
- Local setup and model requirements are not suitable for most casual users.
- The official research workflow should not be confused with every Manycore consumer product.
- Indoor specialization limits claims about arbitrary outdoor or open-ended worlds.
Indoor scene generation and reconstruction, not a general real-time video world or consumer game generator.
04 / Access & output
How the model reaches people
Open GitHub repository and model artifacts for technical users; related hosted generation capabilities are exposed through Manycore's Aholo platform.
- Platforms
- GitHub · Hugging Face artifacts · Aholo API and SDK
- Outputs
- Navigable indoor 3D scene · 3D Gaussian scene representation
- Inputs & compatibility
- Reference image · 3D semantic layout · Text-to-layout workflow · Local CUDA setup
05 / Practical path
How to evaluate it
- 01
Review official examples to confirm that the indoor-scene scope matches the job.
- 02
Use the repository release instructions for local inference and checkpoint selection.
- 03
Inspect room layout and novel views rather than judging only the reference camera.
- 04
Use Aholo documentation when the requirement is a hosted API or SDK.
Primary evidence