Skip to content

PlayCanvas Engine integrates Gracia 4D Gaussian Splatting on the web (Will Eastcott announcement)

Will Eastcott (creator of PlayCanvas and SuperSplat) announces that Gracia’s 4D Gaussian Splatting demos are now integrated with the open-source PlayCanvas engine. The integration brings real-time relighting, shadow casting, post effects, and the ability to mix Gracia’s 4DGS content with other splats and meshes — all running in the browser. This is a tooling/product announcement; no underlying paper.

  • PlayCanvas now supports loading and rendering Gracia’s 4D Gaussian Splats in the web engine [tweet].
  • The integration enables real-time relighting of 4DGS content, shadow casting from splats onto meshes (and vice versa), and post-processing effects [80.lv writeup].
  • 4DGS content can be mixed in the same scene with conventional 3D Gaussian splats and triangle meshes [80.lv writeup].
  • The PlayCanvas/Gracia demo and source code are publicly available [80.lv writeup].

Gracia is a 4D Gaussian Splatting pipeline that streams volumetric video like YouTube clips (Gracia previously shipped 4DGS demos that can be streamed to standalone XR headsets). PlayCanvas is an open-source WebGPU/WebGL engine; SuperSplat is the companion splat-editing toolchain. The integration adds a loader and renderer for Gracia’s 4DGS format inside PlayCanvas, plus hooks to PlayCanvas’s existing PBR lighting and post pipeline so dynamic lights affect the splat field and the splat field contributes to shadows / global look. The tweet links a demo and source code; no architectural details beyond that are claimed.

No quantitative benchmarks are reported. The deliverables are: (a) a runnable web demo of Gracia 4DGS content rendered in PlayCanvas, (b) public source code for the demo, and (c) a feature path for mixing 4DGS, 3DGS, and meshes in one scene with shared lighting.

4DGS-as-streamable-volumetric-video is one of the few paths to photoreal dynamic content delivered through a browser without a custom viewer — directly relevant to anything Luma ships on the web. Complements NeoVerse: Enhancing 4D World Model with in-the-wild Monocular Videos, which uses generated 4D Gaussian fields with bidirectional motion features as the world-model output (Gracia is on the capture/playback end of the same representation). Contrasts with VerseCrafter: Dynamic Realistic Video World Model with 4D Geometric Control, which uses static 3D Gaussians for object-motion control in a video diffusion model — Gracia keeps everything per-Gaussian-per-frame and trades that extra storage for native temporal fidelity. The relight-and-mix story is the new bit: previous 4DGS demos rendered the splat field in isolation, with baked lighting that couldn’t interact with engine content.