Optimize a 3D model

It merges vertices that sit within a set distance of each other, closing tiny gaps left by how the model was exported. weld defaults to 0.0001 in the models own units, and setting it to 0 turns welding off entirely.

It removes triangles toward a target fraction of the original count using three.js SimplifyModifier, so 0.5 aims to keep about half the triangles. A value of 1, the default, leaves the mesh unsimplified; the accepted range is 0.01 to 1.

No. It snaps every vertex position onto one of 2 to the power of N evenly spaced steps across the geometrys bounding box, a plain float32 rounding done before export. glTFs own KHR_mesh_quantization extension changes the accessors component type to a normalized integer instead, and this tool does not write that extension.

Because three.js ships Dracos decoder but not an encoder, and the encoder three.js own exporter points at lives on a CDN this project never fetches anything from. Adding Draco output would mean shipping a dependency this codebase does not otherwise carry.

glb, gltf, obj, stl, stl-ascii, ply or usdz, glb by default. The result is named after your input file with an -optimized suffix and the extension matching your chosen format.

Yes. It takes a single file and has no browser-only restriction, so the CLI and MCP surfaces run the same weld, simplify and quantize code as the page.