visionary

Why we included this project

Visionary attacks the setup problem in neural rendering: instead of installing a Python environment with CUDA toolkits to view a 3D Gaussian Splatting capture, you open a browser tab. The renderer is built on WebGPU with ONNX Runtime handling inference per frame, so it composites millions of gaussian primitives live on your GPU. A three.js plugin with a compact TypeScript API makes it straightforward to embed that rendering into an existing web app rather than treating the browser as a demo shell. The unified asset loader takes PLY, SPLAT, GLB, and FBX files, and the per-frame ONNX inference path lets you run your own trained models for dynamic avatars or generative post-processing. If your scenes mix gaussian clouds with regular meshes, the depth compositing avoids the z-fighting artifacts that usually appear when the two are rendered together.