Drag to look. Move with WASD or the thumbstick.
A pretrained image model imagines the scene. A second neural network reconstructs a persistent 3D Gaussian field. Your device renders that memory as you move.
New imagery is generated on the connected RTX 5080. Saved worlds work without it. Sound and subtle foliage motion are procedural. This is a hybrid prototype with limited movement around each view, not Google's Genie or a full physics world model.
Transform synthesizes a new version. Explore onward uses your current view as a visual starting point; it may change details. Return restores the previous place.