Volumetric Video vs Gaussian Splatting

Capture Workflow and Accessibility

The single largest practical difference between these technologies is the barrier to entry. Volumetric video has historically required purpose-built capture stages costing millions of dollars, staffed by specialized technicians. Companies like Metastage and Dimension Studio operate these facilities as a service, but access remains limited to well-funded productions. AI-driven sparse-camera approaches are lowering the bar — Nokia's standards-based real-time volumetric communication system and 8i's mobile capture solutions both demonstrate that volumetric capture is moving beyond the studio — but the infrastructure remains substantially heavier than Gaussian splatting's requirements.

Gaussian splatting, by contrast, can reconstruct a scene from a set of photographs taken with a consumer camera or smartphone. DJI's integration of splatting into its Terra drone mapping software and Zillow's deployment in SkyTours real estate listings both illustrate how the technology meets users where they already are. For spatial computing developers who need to digitize real environments, splatting has become the pragmatic default.

Rendering Performance and Platform Reach

Gaussian splatting's architectural advantage in rendering speed is difficult to overstate. Because splats are rasterized via GPU-friendly projection and alpha compositing — rather than requiring per-ray neural network evaluation as in NeRF — frame rates of 100+ FPS are routine on desktop hardware. The Mobile-GS system demonstrated 116 FPS on mobile devices in 2025, with scene files compressed to just 4.8 MB. Combined with WebGPU browser rendering, this makes Gaussian splat scenes nearly as accessible as photographs on the open web.

Volumetric video playback, particularly for mesh-based sequences, demands more from the rendering pipeline. Each frame is a distinct 3D mesh that must be decoded, uploaded to the GPU, and rendered — a workflow that strains both bandwidth and compute. Neural compression codecs are improving streaming efficiency, but volumetric video remains a heavier lift for consumer devices, especially in mixed reality headsets where thermal and power budgets are tight.

Dynamic Content and Temporal Performance

This is where volumetric video retains its clearest advantage. It was designed from the ground up to capture motion — human performances, athletic events, live action. The output is inherently temporal: a sequence of frames representing continuous movement. For applications like immersive sports broadcasting, telepresence with life-sized 3D avatars, and concert replays, volumetric video provides a mature, proven pipeline.

Dynamic Gaussian splatting (4D splatting) is closing the gap rapidly. The use of dynamic splats in Superman (2025) marked a watershed moment for the technique in visual effects production. Research from SIGGRAPH Asia 2025 on disentangled Gaussian splatting demonstrated high-fidelity relightable volumetric video through geometry-appearance decoupling. However, 4D splatting workflows remain less mature than traditional volumetric pipelines for long-duration, multi-actor performances.

Visual Fidelity and Scene Complexity

Gaussian splatting excels at reproducing fine geometric detail and view-dependent appearance effects. Hair, foliage, fences, glass, and reflective surfaces — elements that frequently defeat mesh-based reconstruction — render convincingly as splats. The continuous, soft nature of Gaussian primitives naturally handles semi-transparent and fuzzy boundaries that would require extremely dense meshes to approximate.

Volumetric video's visual quality for human subjects is high when captured in professional studios with dense camera arrays, but the mesh reconstruction step can introduce artifacts on fine details. The advantage of mesh-based output, however, is compatibility with established compositing and lighting pipelines — artists can relight, reshade, and integrate mesh-based volumetric captures into scenes using familiar tools. The disentangled splatting research is bringing similar capabilities to Gaussian representations, but tooling maturity still favors meshes for post-production work.

Standardization and Ecosystem Maturity

Both technologies reached important standardization milestones in 2025. Volumetric video benefits from MPEG's V3C (Visual Volumetric Video-based Coding) and MIV (MPEG Immersive Video) standards, which provide a framework for compression, streaming, and interoperability. These are backed by major players like Nokia and Microsoft.

Gaussian splatting gained its own standardization anchor when the Khronos Group released a candidate glTF extension (KHR_gaussian_splatting) for storing splats in glTF 2.0 files. With glTF already the de facto interchange format for real-time 3D, this positions splats for broad tool and engine adoption. Esri's ArcGIS Pro, The Foundry's Nuke 17.0, and 3DVista's virtual tour platform all shipped native splat support in 2025, signaling that the ecosystem is maturing at an extraordinary pace.

The Convergence Path

The most significant trend in 2025–2026 is convergence. 4DViews — one of the longest-running volumetric capture studios — announced Gaussian splatting output at SIGGRAPH 2025, effectively acknowledging that splats are becoming the preferred rendering representation even for studio-captured volumetric content. Generative AI is accelerating this convergence: models can now synthesize Gaussian splat scenes from text or single images, while neural volumetric video systems can reconstruct 3D performances from monocular input.

The future likely involves volumetric capture workflows (multi-camera rigs, depth sensors, LiDAR) feeding into Gaussian splatting representations for delivery and rendering. The capture infrastructure of volumetric video and the rendering efficiency of Gaussian splatting are complementary, not competing — and the industry is building pipelines that combine both.