Computer Vision / Real-time 3D
Gaussian Splatting is a recent technique for 3D scene representation and rendering from images. Instead of traditional meshes, the scene is represented with thousands or millions of 3D gaussians: small volumetric elements containing position, color, scale and orientation. During rendering, these gaussians are projected into screen space and blended to reconstruct the final image efficiently.
The demos on this page are experimental Gaussian Splatting implementations optimized for the web. Bringing this technology to the browser makes 3D scenes easily accessible without installing dedicated software, resulting in an immediate experience across many platforms.
Model quality in these demos is optimized to run on most devices, but it can be adjusted and improved based on target needs and available hardware.
5 million splats scene optimized for desktop
A dense high-volume splat scene representing a natural park centered around a tree stump. This setup shows how complex environments can be represented with rich detail while preserving smooth performance.
400k splats scene optimized for phones and tablets
A balanced experience between visual quality and performance, designed for mobile devices. It keeps fluid rendering even on constrained hardware thanks to advanced optimization techniques.
