PlayCanvas updated its engine and converter to support the compressed LCC2 format for splat streaming, improving scene delivery on many devices.

Engine update

The update introduces a high-performance WebGPU handler and level-of-detail streaming to serve high-quality scenes on desktop and mobile devices. Earlier, the team added a collision generator so uploaded splats provide walkable, non-penetrating geometry for users.

WebGPU compute pipeline

Instead of sorting splats inside a worker thread, the new WebGPU handler shifts heavy tasks to compute shaders on the GPU for culling then projection. Remaining splats are sorted using a fast radix sort on the GPU, while final rasterization still uses vertex and fragment shaders that operate on already-projected data.

Streamed SOG and LOD

SuperSplat now automatically converts uploaded splats into Streamed SOG, extracting LODs and serving scene pieces on demand to the viewport. The system loads a low LOD first for immediate display, then progressively fetches higher-detail tiles, enabling much larger scenes on phones than before.

Compatibility and fallback

WebGPU is reported to be available on ~84% of devices (recommended via Chrome), and SuperSplat automatically falls back to rendering with WebGL2 when necessary.

Sample splat scenes

• Railway museum, Freilassing, Germany
• Shrine from XV–XVI centuries, Casnigo commune, Italy
• Primeval forest by Sirakoma pond, Japan
• Strawberry (small scene)
• Arti-2026 exhibition
• Mononoke Forest, Japan
• Stonehenge
• Els-Vilars fortress, Spain
• Vuores, Finland
• Sunset in an ice cave (image sourced from ChatGPT)
• 3D printer motherboard
• Lublin, Poland — 250 million splats (6 GB)
• Interstellar-style scene — 10 billion splats

Long-open tabs may consume much memory: after opening Lublin and the Interstellar scene, reported VRAM grew above 16 GB within thirty minutes, suggesting a possible memory leak that developers may address.