PlayCanvas Open Sources SOG Format for Gaussian Splatting (blog.playcanvas.com)

PlayCanvas has open sourced SOG (Spatially Ordered Gaussians), the next-generation, super-compressed format for 3D Gaussian splats that replaces SOGS. SOG can compress massive scans dramatically—PlayCanvas shows a 4M‑gaussian skate park reduced from a 1 GB PLY to a 42 MB SOG (~95% savings)—while improving load speed and fidelity. The company published the spec plus reference implementations: SplatTransform (writer/CLI) and loader/rendering in the PlayCanvas Engine (support added in v2.11.0), and the PlayCanvas Editor now accepts bundled .sog assets. SuperSplat has also added SOG support, making compressed scans faster to stream and practical on memory‑constrained devices. Technically, SOG keeps the meta.json + .webp image approach but reorganizes splat data in Morton order so files are "GPU‑ready" and require no CPU preprocessing at load time. Compression is more efficient (smarter bit allocation) and platform-agnostic: unlike SOGS, SOG’s pipeline runs with WebGPU instead of CUDA, widening device compatibility. Developers can create .sog files with the open source SplatTransform CLI (npm install -g @playcanvas/splat-transform; splat-transform input.ply output.sog). By open sourcing the format and reference code, PlayCanvas invites engine and tooling developers to adopt and contribute to an interoperable, high-performance standard for 3D Gaussian splats.