Palatial opened scene reconstruction in closed beta on November 12, turning photos, walkthrough videos and scans of indoor spaces into robot simulation scenes.
A scene can start from a description, a few photos, a walkthrough video, a 360 capture, a floor plan or a scan. Palatial lists egocentric videos, Gaussian splats and large scale walkthroughs among its sources, and names video walkthroughs, point clouds or splats, and photos from several angles as the best inputs today. Palatial's earlier releases centered on single objects, including the Mad Max agent for articulated SimReady assets in September and an open simulation asset library in October. Scene objects now come from that asset generation pipeline, the open library and a workspace's own assets.
Each scene is built by a scene director, an agent that plans the work, checks its own output and works in drafts. Early drafts block out walls, openings and large furniture, and later drafts add smaller objects, materials and lights. Every draft is rendered from the same saved views, and reconstruction drafts are shown beside the original frames. Feedback goes through an agent chat, and the director can fill in parts of a room the footage never covered.
The output is USD, with Omniverse based sensors and rendering through NVIDIA's ovrtx, which gained Gaussian splat support in its 0.2.0 release. A delivery is a full detail USD scene plus the individual SimReady assets inside it, with collision geometry, physical properties and working joints, so drawers slide, doors swing and cabinets open. Repeated objects are instanced, and background props are simplified. Palatial does not state whether a splat supplied as input is carried into the delivered scene or converted to geometry.
Indoor spaces such as homes, kitchens, offices, labs and stores are the current focus, and outdoor scenes are less tuned. Without a size reference such as printed markers, known dimensions or a floor plan, scale is estimated. Anything the director fills in is, in Palatial's wording, "plausible rather than measured," and each scene records which parts were observed and which were generated. Large scenes can take several hours to generate.
Approved workspaces receive 100 additional tokens. Access to the closed beta can be requested from the Palatial dashboard.





