Khronos Announces glTF Gaussian Splatting Extension

Cookie Policy

Clicking 'Accept All Cookies' means you agree to our cookie use and data sharing as described in our privacy policy under 'Cookies and Tracking Technologies'.

Manage cookie options

Skip to main content

Press Release

Khronos Announces glTF Gaussian Splatting Extension

Cross-platform baseline for storing 3D Gaussian splats in glTF while enabling future innovation; Community feedback invited before ratification

Beaverton, OR – February 3, 2026 – Today, The Khronos® Group, an open consortium of industry-leading companies creating graphics and spatial computing standards, announces a release candidate for the KHR_gaussian_splatting baseline extension. This extension enables storing 3D Gaussian splats in glTF® 2.0, the most widely adopted 3D asset delivery format. A release candidate allows for broad industry feedback before ratification to ensure the final specification meets industry needs.

The new extension establishes an open foundation for representing Gaussian splats, an increasingly important geometry and radiance field representation, across real-time graphics, digital twins, and large-scale geospatial visualization pipelines.

“KHR_gaussian_splatting marks a major milestone for glTF, extending the format to support an entirely new class of geometric representation,” said Neil Trevett , president of the Khronos Group. “By bringing the Gaussian splatting community together around a standards-based approach, Khronos is helping ensure this powerful new rendering technique can scale across tools, platforms, and the web.”

Gaussian Splatting: Real-Time Radiance Field Rendering

Gaussian splatting is a radiance field representation technique that converts multiple 2D images into a photorealistic 3D asset. Photos or videos are used to create a sparse 3D point cloud of an object or scene, with each point defined by properties including position, scale, rotation, color, and opacity. Machine learning optimizes these Gaussian points to match the original input images, and the optimized points are projected onto a 2D surface via rasterization to produce highly responsive camera views.

3D Gaussian splat tileset of a chemical refinery embedded with Cesium World Terrain in CesiumJS. Data from Bentley Systems.

This approach has proven especially well-suited to geospatial capture workflows, including:

City-scale and corridor-scale reality capture

Complex natural environments with vegetation and irregular geometry

Reflective, translucent, or detail-rich urban surfaces

Rapid field acquisition using commodity cameras, drones, or mobile devices

3D Gaussian splat of a power substation embedded alongside Google Photorealistic 3D Tiles in CesiumJS. Data from Bentley Systems.

Gaussian splats can be trained quickly, achieve high frame rates, and scale from individual objects to entire urban environments—making them increasingly attractive for mapping, digital twins, infrastructure monitoring, simulation, and situational awareness.

Gaussian splatting is also being quickly adopted in many other markets including photojournalism, media and entertainment, robotics training, cultural preservation, and is poised to bring 3D capture and display to social media. At the same time, generation, training, rendering, and compression techniques continue to evolve and improve.

Industry Need for Standardization

Without standardization, this rapid evolution could easily lead to fragmentation. In January 2025, the Metaverse Standards Forum initiated a series of public Town Halls to explore whether Gaussian splatting was ready for standardization. Stakeholders identified strong overlap across use cases and interoperability challenges that could be meaningfully addressed through open standards—specifically by enabling splat storage in glTF assets.

A key conclusion was the importance of enabling Gaussian splats to integrate cleanly into existing spatial data ecosystems, where assets must coexist with meshes, terrain, imagery, and sensor-derived data, often within Earth-referenced coordinate systems. Standardizing Gaussian splat storage within glTF was identified as a critical step toward that goal.

The Khronos 3D Formats Working Group responded with a focused development effort and is now gathering a final round of industry input on the resulting KHR_gaussian_splatting glTF extension. Khronos encourages engine developers, implementers, creators, and artists to explore the specification, experiment with sample assets, develop implementations, and share their feedback. Strong community input will help ensure a solid foundation and vocabulary for interoperable, spatially contextualized 3D Gaussian splat content delivered in the glTF format while supporting ongoing innovation.

3D Gaussian splat of Kansas City Coffee Roaster courtesy of ShareUAV collected with a ShareUAV 102S sensor (2,431 images) in an area survey pattern. This camera has 5 cameras (1 nadir, 4...