Research
We are a frontier AI lab building world models that understand space, not just pixels. Our latest work, Echo-2, is generally available in private preview.
Echo-2 reconstructs physically-grounded geometry, consistent lighting, and material properties from any text prompt, image, or panorama — producing worlds you can walk through, not just look at.
We introduce a depth-consistent diffusion process that lifts single 360° captures into navigable 6-DoF meshes while preserving material and lighting cues.
A study of collision, gravity, and scale consistency in generated 3D spaces, and how grounding improves downstream simulation and robotics transfer.
spacial is in private preview. Join the waitlist to start building navigable worlds.