GAUSSIAN SPLATS

Gaussian Splats represent 3D scenes as millions of tiny semi-transparent ellipsoidal "blobs," each storing position, shape, color, and opacity, rather than traditional polygon meshes. This workflow-compressing technology offers 
Artists and Studios an opportunity to add this powerful tool to existing pipelines.

Why use Gaussian Splats?

Speed advantage | Gaussian Splats rasterize directly on the GPU, enabling real-time 60fps+ rendering. 

No painful cleanup | Traditional photogrammetry requires hours of retopology, UV unwrapping, and texture baking. GSplats bypass all of that overhead. 

Hard materials solved | Hair, fur, foliage, and wet surfaces which are historically the hardest things to model are captured naturally and look photorealistic without special treatment.

Virtual production | Studios can scan real locations and drop them into Houdini and Unreal Engine as backgrounds or lighting references, skipping expensive set builds or location travel. 

VFX compositing | GSplats give artists accurate 3D geometry for match-moving, depth occlusion, and lighting reference when blending CG elements with live footage.


CITY | 23 M Points | 5 GB

How to Capture Gaussian Splats

Gaussian Splats can be trained in Houdini using TOPs after external camera solving, with monitoring, procedural cleanup, and new Copernicus rasterization tools. This enables 2D compositing, depth extraction, and image-based editing while preserving high-fidelity scanned detail.

Take Photos | After external camera solving, PDG/TOPs trains Gaussian Splats in Houdini using PDG/TOPS parallelism for efficient, scalable processing across multiple machines.

PDG/TOPS |  Accelerate Gaussian Splat training on render farms while monitoring tools visualize machine learning progress and progressively improve reconstruction quality.

Cleanup | Houdini enables procedural G-splat cleanup, removing peripheral artifacts while preserving high-fidelity lighting, reflections, and fine surface detail.

Rasterize | Copernicus adds Camera COP and G-splat rasterization, converting 3D splats into 2D images with color and depth for compositing.

Houdini-generated GSplats

Houdini generates Gaussian Splats directly from Karma renders using known camera and scene data, preserving attributes for high-fidelity, real-time visualization of meshes, volumes, and effects. The resulting lightweight assets enable reusable, interactive rendering, efficient memory usage, and compelling proxies for set dressing and real-time workflows.

Karma | Houdini converts Karma-rendered scenes into Gaussian Splats by capturing multiple known viewpoints with camera arrays, eliminating external camera estimation.

Captured Info | Rendering-generated Gaussian Splats preserve scene attributes like positions and normals, enabling richer data capture and workflows beyond photographic reconstruction.

CAT | 23 M Points | 5 GB     

CLOUD | 494 K Points | 113 MB

EXPLOSION | 1.9 M Points | 440 MB

GSplats in Motion

Because Gaussian Splats are fundamentally points with attributes, they integrate seamlessly into Houdini’s procedural workflows. Standard deformation nodes, including Bone Deform and Surface Deform, animate splats without specialized tools. While simulations can directly drive splat points to enable realistic character animation and dynamic effects.

MOTION | Gaussian Splats support simulations, allowing effects like realistically wind-animated vegetation to drive splat points with simulation data.

RIG & ANIMATE | Animated Gaussian Splats integrate with standard Houdini deformation tools, enabling character animation using Bone Deform and Surface Deform workflows.

Rendering GSplats

SideFX Labs introduces prototype Gaussian Splat relighting tools that remove baked lighting, regenerate illumination using normals and Copernicus, and leverage spherical harmonics for convincing results. Houdini 22 also improves Karma rendering quality, advances Vulkan real-time rendering, and enables interactive HDRI relighting and compositing workflows, expanding creative possibilities for splat-based content.

COPERNICUS | Gaussian Splats support HDRI relighting and interactive Copernicus compositing, enabling real-time effects like chromatic aberration while exploring new workflows.

RELIGHT | SideFX Labs prototypes Gaussian Splat relighting using Copernicus, normals, and spherical harmonics to overcome baked-lighting limitations with convincing dynamic illumination.