Gaussian Splatting for Unity, with a focus on Mixed Reality. Import and render trained 3DGS and 2DGS models, or animate supported 4DGS models through the 3D renderer. Includes single-pass instanced stereo rendering with foveation support.
This adds performance improvements targeting XR, 2DGS and 4DGS support, instanced and foveated rendering, among others. Now distributed as a Unity package that can directly be installed and used.
NOTE:
This is a research project building on UnityGaussianSplatting. No support.
Requires Unity 6000.3 or later and D3D12, Metal or Vulkan. Burst, Collections and Mathematics are installed as package dependencies. URP/HDRP integrations use Unity 6 render pipelines; 4D import and playback additionally require Unity AI Inference 2.6.x.
In the Unity Package Manager, choose Add package from git URL and enter:
https://github.com/ninjamode/Unity-VR-Gaussian-Splatting.git
For a local checkout, choose Add package from disk and select this folder's package.json.
- Open Tools → Gaussians → 3D → Create Splat Asset, select your trained Gaussian file, choose a quality preset and click Create Asset.
- Add Gaussians → 3D Splat Renderer to a GameObject and assign the generated asset.
- Configure the render pipeline. For URP, add Gaussian Splat 3D URP Feature to the camera's renderer asset and keep Render Graph enabled. For HDRP, add the matching Gaussian Splat 3D/2D HDRP Pass to a Custom Pass Volume before transparent rendering.
Use the 3DGS guide for further info. Keep generated data files with their assets.
Device Support: Works on desktop as well as Quest 3. Apple Vision Pro should be supported but is untested. Other VR systems should be fine too.
Built-in doesnt need any special configuration, but is slower. HDRP is mostly untested.
Same as with 3DGS, create 2D asset, add 2D Splat Renderer, add the Gaussian Splat 2D URP Feature or HDRP pass to the camera asset.
A little export helper is necessary to have all the data for runtime use in unity. Supported models use the HexPlane + MLP architecture from hustvl/4DGaussians.
Export your trained checkpoint with the offline exporter, then copy its whole output folder to your Unity machine. With Unity AI Inference installed, use Tools → Gaussians → 4D → Create Splat Asset.
Assign <model>-canonical3d.asset to a 3D Splat Renderer, add 4D Gaussian Deformation to the same GameObject and assign the matching <model>-inference.asset. The canonical model can render without inference. See the 4DGS guide for import and playback.
- WebGL and OpenGL graphics APIs are not supported.
- Models are sorted individually. Overlapping renderer objects do not share a global per-splat order.
- Lower sort precision and size/opacity pruning trade visual quality for reduced work. Check output on the intended device and pipeline.
- Selection editing can be flaky.
The 3D implementation is derived from Aras Pranckevičius' UnityGaussianSplatting. Thank you! The 2D renderer adapts the same foundation with different loading and display logic.
Package code is available under the MIT license. See third-party notices for included software. Trained models and datasets retain their own licenses. AIs have been used for coding support, bad ideas are my own.