juunini/gltf-optimizer

Optimize glTF. Draco compress and convert texture to WebP.

GLTF Optimizer

---

Demo

Caution

Only can compress .glb file now.
Will be support .gltf soon.

Introduce

Optimize glTF.
Draco compress and convert texture to WebP.

Using glTF-Transform and webp-converter-browser

Install

# If you want to use CLI
npm install --global gltf-optimizer

# npm
npm install gltf-optimizer

# yarn
yarn add gltf-optimizer

# pnpm
pnpm add gltf-optimizer

# bun
bun add gltf-optimizer

# If you install and use browser side
cp node_modules/gltf-optimizer/draco3d/* ./<your_static_path>/

Usage

import { optimizer } from 'gltf-optimizer'

// ...

// node (backend side)
const glb = fs.readFileSync('./target.glb')
const optimized = await optimizer.node(glb, { /* options */ })
fs.writeFileSync('./compressed.glb', optimized)

// browser (frontend side)
const optimized = await optimizer.web(glb, { /* options */ })
// if using get-file-using-a-tag (https://github.com/juunini/get-file-using-a-tag)
download({ fileName: 'compressed.glb', arrayBuffer: optimized })

Usage(CLI)

gltf-optimizer -i model.glb
gltf-optimizer -i model.glb -o ./output

Command-Line Flags

Flag	Description	Required
--help, -h	Display help	No
--input, -i	Path to the glTF or glb file.	✅ Yes
--output, -o	Output path of the glTF or glb file. Separate resources will be saved to the same directory.	No, default ./
--emissiveStrength	Emissive strength of the glTF file.	No, default 1.0
--draco.method	edgebreaker or sequential	No, default edgebreaker
--weld.tolerance	Tolerance, as a fraction of primitive AABB, used when merging similar vertices.	No, default 0.0001
--simplify.enabled	Enable/disable vertex simplification.	No, default true
--simplify.ratio	Target ratio (0-1) of vertices to keep.	No, default 0.75
--simplify.error	Limit on error, as a fraction of mesh radius.	No, default 0.01
--texture.resize.resolution	Maximum width/height to enforce, preserving aspect ratio. For example, a 4096x8192 texture, resized with limit [2048, 2048] will be reduced to 1024x2048.	No, default 1024
--texture.resize.filter	Resampling filter method. LANCZOS3 is sharper, LANCZOS2 is smoother.	No, default LANCZOS3