WilliamLiu-1997/Gaussian-Splat-Lite

High performance 3D Gaussian Splatting (3DGS) renderer for Three.js. WebGPU and WebGL2. Supports PLY, SPZ, Streaming LOD including SOG, RAD and LCC2.

What it solves

Rendering 3D Gaussian Splatting (3DGS) models in a web browser is computationally heavy and often requires specialized code. This library integrates with Three.js, letting developers use familiar scene and camera APIs while achieving high performance through WebGPU or WebGL2. It also solves the problem of very large scenes by streaming only the detail currently visible, instead of loading everything at once.

How it works

You add a SplatMesh object to a Three.js scene and a GaussianSplatRenderer to handle all visible splat models. The renderer works with either WebGPU (the default) or WebGL2, and you can force WebGL2 if needed. For large scenes, you use a streaming scheduler (RAD, SOG, or LCC2) that loads levels of detail as the camera moves, with smooth transitions between them. The renderer supports stochastic rendering for transparency, optional neural denoising, offscreen capture to images or environment maps, and SDF-based edits to recolor or hide parts without moving the splats.

Who it’s for

Web developers and Three.js users who need to display interactive 3D Gaussian Splatting scenes—especially those with very large models that must stream efficiently. It’s also useful for anyone requiring WebGPU/WebGL2 compatibility, environment reflections, or the ability to edit splat models programmatically.

Highlights

  • Works with both WebGPU and WebGL2 using the same Three.js scene API
  • Streams RAD, SOG, and LCC2 scenes with adaptive levels of detail
  • Optional stochastic rendering with an experimental neural denoiser
  • Offscreen capture to images, cube maps, and environment maps
  • SDF-based editing to recolor or hide parts of a model
  • Loads PLY, SPZ, SOG, and RAD files from URLs, files, or bytes
  • Supports large GIS/ECEF scenes via accurate placement

Related

  • Project
  • Project
  • Project
  • Project