SamWORLD — Shroozy

SamWORLD turns a room you already know into a world you can explore. Film a short walk around your room, upload the video or photos, then type a world: a haunted Victorian library, an underwater coral cave, or a retro arcade. The application reconstructs the space, identifies its furniture, and uses the prompt to restyle surfaces and objects while keeping their relationship to the original room. Built by Tiwa Bakre at Shroozy, the project brings AI world creation to people who do not use 3D modelling tools. Instead of starting with a blank scene, you start with your own surroundings. The goal is a creative space you can explore on a screen and, through the Meta Immersive Web SDK/WebXR integration, step inside on Quest. SamWORLD could help architects and home designers explore different looks and atmospheres for a space, or simply let anyone have fun experiencing their own home in a completely different way. The implementation combines Meta MapAnything reconstruction, Muse reasoning, SAM 3.1 segmentation, Muse Image texture and object restyling, and SAM 3D object reconstruction. The backend stores scans and restyles, caches expensive results, tracks progress per element, and applies bounded concurrency and model-call budgets. The viewer supports the original scan, reconstructed room surfaces, object selection, and progressive restyling. This is a hackathon prototype: reconstruction quality depends on capture coverage, missing geometry may be estimated, and generation can take time. The video distinguishes explanatory animation from captured product behaviour. Live app: https://frontend-production-9b6d.up.railway.app/ A note on the demo: I put so much time into polishing SamWORLD itself that I wasn’t able to create the demo video I’d originally imagined.