
Link / Media
A step-by-step, fully open-source pipeline that turns phone photos or video of a real space into a photoreal, walkable 3D scene running in a browser tab, no app, no VR headset required.
Built entirely on PlayCanvas's open-source tools, this guide walks through the complete process of turning a phone photo dump into a navigable 3D scene using Gaussian splatting.
Capture: 50 to 200 overlapping photos or a slow walking video of the space, with guidance on lighting, height variation, and what to avoid (moving objects, reflective surfaces).
Reconstruction: either a fast path through Polycam or Luma AI, which handle reconstruction in the cloud, or a full DIY route using COLMAP and Gaussian splatting for anyone with (or renting) a CUDA GPU.
Cleanup and conversion: SuperSplat for trimming and reorienting the raw scan, and splat-transform for compressing it into a fast-loading web format.
Publishing: three paths depending on how much control you want, instant hosting through SuperSplat, self-hosting via supersplat-viewer, or building a fully custom interactive app in the PlayCanvas Editor with multi-room navigation and branding.
Includes practical notes on where the technique breaks down (glass, reflections, poor lighting, moving objects) and a full toolchain reference table linking every tool used at each step.