SuperSplat: Browser-Based 3DGS Editing and OpenUSD Integration

The Pitch

SuperSplat is an MIT-licensed, browser-based editor for 3D Gaussian Splatting (3DGS) that recently achieved integration into OpenUSD and Khronos glTF standards (The Future 3D). Developed by the PlayCanvas team, it enables the editing and publishing of neural captures directly in the web runtime without requiring local desktop software (GitHub). It is currently gaining traction on Hacker News due to its ability to maintain 100+ FPS on consumer-grade hardware while handling multi-million point datasets (PlayCanvas News).

Under the Hood

The February 2026 launch of "SuperSplat Studio" shifted the tool from a simple editor to a suite for building guided narratives and hotspots (RadianceFields.com). A critical technical milestone is the "SOG" (Splat Optimized Graphics) format, which uses WebP-style compression to achieve file sizes 20x smaller than raw splat data (PlayCanvas Blog). This compression is vital for web delivery, as standard 3DGS files frequently exceed 1.5GB, posing significant storage and bandwidth overhead (Medium).

Navigation within the browser is managed through "Walk Mode" and "Streamed LOD," which prevent memory heap overflows during the visualization of large scenes (PlayCanvas News). However, hardware limitations remain a bottleneck; performance degrades significantly on devices lacking modern WebGPU or WebGL2 support (GitHub Issue #132). It is also worth noting that SuperSplat is not a tool for precision engineering, as it lacks the sub-centimeter accuracy found in traditional SfM-MVS workflows (Copernicus.org).

We don't know yet whether an enterprise pricing model exists for the Studio features, as the repository remains listed as free and open-source (GitHub). Furthermore, the specific hardware used to achieve the fidelity seen in the recent "Strawberry" macro demo has not been made public (HN). Most developers will find the current feature set sufficient for high-fidelity web visualization, provided they account for the initial 500MB+ download sizes.

Marcus's Take

SuperSplat is the only viable choice for teams needing to deploy high-fidelity neural captures to the web without locking users into proprietary viewers. The adoption of OpenUSD and glTF as of April 2026 suggests the industry is finally moving past experimental silos and into standardized production pipelines. Use it for marketing, real estate tours, or digital twins where visual "feel" matters more than millimetric surveying. It is a solid piece of kit that won't bloat your tech stack with unnecessary desktop dependencies.

Ship clean code,
Marcus.