Describe the situation your model gets wrong. Studio builds it from 10,000+ environments and 100,000+ assets, varies it thousands of ways, and writes exact ground truth onto every frame.
One living room, sixteen variations, one camera path. Rendered by Studio.
My robot misses small objects on the floor when the sun hits the living room.
You write the request the way you would brief a colleague. Studio does the 3D work, and nothing renders until you approve it.
A warehouse aisle, a nursery, a kitchen at sunset. Studio finds it in the library.
Clutter, layout, light, camera. Whatever you leave out is held still.
Depth, normals, segmentation, boxes and camera pose, written onto every frame.
Preview the variants, then send the final batch to the render farm.
Depth, normals and segmentation are written by the renderer from the 3D scene, not estimated from the picture, so they stay exact in low light, in hard sun and behind clutter.
A nursery, then the same frames with boxes and classes, instances, normals and depth.
A warehouse aisle filmed at robot height, wiped through its layers.
| Image or video | RGB render, up to 1920 × 1080 |
| Geometry | Depth in metres and surface normals |
| Objects | Instance and class segmentation, 2D boxes with class and instance ID |
| Camera | Pose for every frame |
| Scene | Light, layout, object list and a record of every axis that changed |
Homes, warehouses, hospitals, shops, labs and offices, built from real 3D geometry with known materials. Every Studio user works with all of it.

Studio is free for research teams during the beta.