TL;DR:

  • Volumetric video captures a real person or object as a 3D mesh you can walk around — unlike flat 2D video, viewers can change their perspective relative to the subject
  • Production-grade capture still requires specialist studios (arrays of 100+ cameras), but real-time capture for meetings and collaboration is becoming accessible via depth cameras
  • Apple Vision Pro’s Spatial Personas and Microsoft Mesh’s avatar system both use volumetric principles — the gap between “meeting tool” and “true volumetric presence” is closing fast

The fundamental limitation of video calls — even on a spatial headset — is that everyone is still a flat rectangle. You can see someone’s face, but you can’t see them from a different angle, and they have no physical presence in your shared space. Volumetric video is the technology that changes this: instead of capturing a person as pixels on a 2D plane, it captures them as a 3D object you can view from any angle.

The technology has existed in research labs for a decade. In 2026, it’s moved into production applications.

What Volumetric Video Actually Is

Standard video is a time series of 2D images. Volumetric video is a time series of 3D meshes — reconstructed point clouds or polygon meshes that represent the geometry of whatever was captured, along with texture maps applied to the surface.

To view volumetric video, you need a display that can render 3D geometry from an arbitrary viewpoint. This is exactly what VR/MR headsets do natively — they render 3D scenes from the viewer’s position. A volumetric clip of a person sitting at a table, dropped into an XR scene, can be viewed from any angle: from the side, from above, from close up. The captured person has actual spatial presence.

The capture process — at professional quality — requires an array of cameras (typically 80–130 cameras arranged around the subject in a dome or cylinder), synchronized high-speed capture hardware, and significant post-processing to reconstruct the geometry from the multi-camera input. The output files are large: professional volumetric captures typically run at hundreds of gigabytes per minute of footage before optimisation.

Where It’s Being Used Now

Immersive entertainment and live events

The highest-profile application. Volumetric capture is being used for music performances, where an artist is captured once and their 3D likeness is placed in XR environments. Several major touring productions have experimented with this for pre-show or interstitial content. The appeal: a volumetric performance of an artist can be experienced at multiple venues simultaneously, or archived indefinitely.

Sports broadcasting has also adopted volumetric technology for tactical analysis — players are captured volumetrically during training, allowing coaches and analysts to replay and view actions from angles that no camera captured during the original session.

Enterprise training and simulation

Capture a skilled technician performing a complex repair procedure, render it volumetrically, and a trainee can walk around the procedure and view it from any angle — including from underneath or from positions that would be physically impossible for an observer in real life. Johnson Controls and BAE Systems both have pilot programmes using volumetric capture for maintenance training.

Spatial computing collaboration

Microsoft Mesh uses volumetric principles for its “mesh avatars” — reconstructed 3D representations of meeting participants rather than flat video tiles. The reconstruction is done from 2D camera input using real-time neural rendering, not a full multi-camera dome capture.

Apple Vision Pro’s Spatial Personas are similar: a 3D representation of the user, reconstructed from the headset’s cameras and rendered in a shared virtual space. You can see a Spatial Persona from different angles as you move your head. It’s not true volumetric video (it’s a real-time neural mesh reconstruction), but it’s the closest consumer experience to volumetric presence.

Capture Technology in 2026

Professional studio capture (Arcturus, 4DViews, Depthkit)

High-quality productions use specialist volumetric capture studios. The major providers — Arcturus (now part of Snap), 4DViews, Depthkit — operate multi-camera domes in Los Angeles, London, and a handful of other cities. Bookings are expensive (typically £5,000–£15,000 per day for a London studio), but the quality is broadcast-grade.

Output files use formats like holo video (.hvid) or the emerging V-Nova VC-6 volumetric codec, designed for real-time playback at acceptable file sizes.

Depth camera capture (LiDAR + RGB)

For lower-fidelity applications — collaboration tools, enterprise training where photorealism isn’t critical — real-time volumetric reconstruction using depth cameras (Intel RealSense, Azure Kinect’s successor hardware, or LiDAR arrays) is now feasible. The geometry is less detailed than studio capture, but it runs in real time.

Microsoft Azure Remote Rendering and Apple’s Object Capture API both support this workflow for static objects. For animated humans, companies like Avaturn and Ready Player Me are building pipelines that reconstruct animatable avatars from short 2D video inputs, which — while not truly volumetric — achieve a similar presence effect.

NeRF and Gaussian Splatting

Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (covered in detail in our Gaussian splatting guide) can reconstruct volumetric scenes from standard camera arrays, without specialist hardware. For static scenes (environments, objects) the quality is very high. For dynamic capture of people, the rendering latency has historically been prohibitive, but improvements in real-time Gaussian splatting in 2025–2026 have made short volumetric clips of people achievable on consumer hardware.

File Size and Streaming

This is the practical constraint that limits widespread deployment. A 30-second professional volumetric clip at reasonable quality — after compression with modern codecs like V-Nova VC-6 — is typically 500MB–2GB. Streaming this requires fast connections and careful progressive loading strategies.

The practical workarounds:

  • Level-of-detail (LOD) reduction: lower geometry complexity for distant viewpoints, higher for close-up
  • Predictive pre-loading: if the user is in a spatial environment, preload volumetric clips based on where they’re likely to move
  • Hybrid approaches: stream 2D video until the user approaches a threshold distance, switch to volumetric for close-up presence

Web-based delivery of volumetric content is handled by libraries like Depthkit.js and Volograph, which handle codec detection and adaptive bitrate selection for WebXR contexts.

What’s Coming

The trajectory in 2026 points toward convergence between real-time neural reconstruction (Spatial Personas) and pre-captured studio quality. Apple’s research in Gaussian Splatting avatars — extracting a high-quality volumetric model from a few minutes of iPhone video — is likely to appear in a future Vision Pro OS update. When that happens, anyone with an iPhone and a Vision Pro can create and share volumetric representations of themselves without studio access.

For spatial computing developers, the practical takeaway today: Depthkit and 4DViews studio capture are production-ready for pre-recorded content; Microsoft Mesh and Apple Spatial Personas handle real-time presence for enterprise collaboration; and the NeRF/Gaussian Splatting pipeline is worth experimenting with for environment capture and lower-fidelity human representation.

Flat video in XR is a transitional state. The direction of the medium is clear.