WebGPU-Native Runtime
for Next-Generation World Models

A browser-first rendering and inference stack that unifies WebGPU neural rendering with per-frame ONNX Runtime execution,

enabling world-model experiences that are dynamic, extensible, and easy to deploy.

it's free
1
Heterogeneous Assets
ONNX-Driven Per-Frame Inference
Built on WebGPU + ONNX Runtime, Visionary provides a unified runtime for Gaussian Splatting families (3DGS, MLP-based 3DGS, 4DGS, neural avatars) and traditional meshes—running directly in the browser. A standardized Gaussian Generator contract (ONNX I/O + metadata) makes algorithms plug-and-play: models can generate or update Gaussians every frame, and the pipeline can be extended with feed-forward post-processing networks.
1
2
High Performance
WebGPU-First Architecture
Designed around native WebGPU compute + graphics, the engine supports million-scale Gaussian rendering with GPU-side workloads such as parallel primitive sorting. It also provides a hybrid rendering pipeline with depth compositing between Gaussian point clouds and standard meshes to handle occlusion correctly in complex scenes. Universal asset loading across common formats: PLY / SPLAT / KSplat / SPZ / SOG, GLB/GLTF/FBX/OBJ, and ONNX-based dynamic assets (e.g., 4DGS / avatars / scaffold-style representations).
2
3
Plugin-Ready
Seamless Three.js Integration
Visionary ships as a three.js plug-in with a concise TypeScript API, making it straightforward to embed into existing web apps. No heavy native dependencies—share a link, run in-browser, iterate fast, and extend with your own ONNX models.
1