Skip to content

GLB compressor & texture optimizer

Shrink GLB file size with Meshopt geometry compression and WebP texture optimization. Files stay in this browser.

Geometry compression
Meshopt precision
Max texture size
Texture format

Viewport

Open a GLB to view its mesh.

More browser GLB tools

Browser GLB Compression

How to compress a GLB file without losing visual fidelity

Shrink vertex buffers with Meshopt or Quantize and optimize embedded textures in three client-side steps.

  1. 01

    Open a static GLB

    Drop any uncompressed static GLB up to 30 MiB and 250,000 triangles. Compression, cleanup, and texture encoding run locally in a Web Worker with zero uploads.

  2. 02

    Choose Meshopt & texture settings

    Pick Meshopt or Quantize geometry encoding, enable weld/prune/dedup cleanup, and cap embedded textures at 512 px, 1024 px, or 2048 px in WebP, JPEG, or PNG.

  3. 03

    Compare size savings & export

    Inspect original and compressed models side by side, check exact bytes and percentage saved, and download the smaller GLB.

Optimization Pipeline

Target both mesh buffers and embedded images

Combine cleanup passes with modern buffer compression and right-sized textures for faster web and mobile loads.

EXT_meshopt_compression

Quantizes and encodes vertex and index buffers for Three.js, Godot 4, and modern glTF 2.0 loaders without changing triangle count.

Weld, prune & deduplicate

Strips unused accessors, merges duplicate materials or textures, and shares identical vertices cleanly before encoding.

Texture resize (512–2048 px)

Downscales oversized 2K or 4K embedded images inside the browser so texture weight matches real gameplay distance.

WebP re-encoding

Converts PNG and JPEG maps into compact EXT_texture_webp textures while keeping alpha channels and PBR material slots intact.

When to compress vs decimate polygons

Use the GLB compressor when your triangle count is already right and you want smaller network downloads via Meshopt, Quantize, and WebP textures. Read our GLB file-size guide for real byte benchmarks.

Technical Specifications & Guardrails

File limits, runtime guarantees & scope

Supported specifications & hard limits

Input format
Uncompressed static glTF 2.0 binary (.glb)
Maximum file size
30 MiB · up to 250,000 triangles / vertices
Geometry codecs
EXT_meshopt_compression (medium/high), KHR_mesh_quantization, or Cleanup only
Mesh cleanup passes
weld() + dedup() + prune({ keepSolidTextures: true })
Texture optimization
Downscale cap (Original / 2048 / 1024 / 512 px) + WebP / JPEG / PNG
Execution & privacy
100% local Web Worker (@gltf-transform + meshoptimizer) · Free

What GLB Compressor does NOT do

  • Does not output Draco (KHR_draco_mesh_compression) or KTX2/Basis Universal GPU textures—it intentionally uses EXT_meshopt_compression, KHR_mesh_quantization, and EXT_texture_webp for faster WebAssembly decoding and crisp low-poly vertex preservation.
  • Does not reduce triangle count; compression keeps your exact polygon count while shrinking binary buffers and embedded images. Run Polygon Reducer first if your mesh has too many triangles.
  • Does not accept GLBs that already use compression extensions (Draco/Meshopt), external URIs, skins, animations, or morph targets.
  • Does not upscale small textures—the 512/1024/2048 px cap only shrinks images larger than the selected maximum dimension.
Real Measured Compression Benchmarks

Actual byte savings on bundled CC0 GLB samples (zero triangles lost)

Measured directly with our production compressGlb worker (cleanup enabled: weld + dedup + prune, Meshopt medium vs Quantize). Every model retains 100% of its triangles, materials, and meter bounds.

Sample GLBTriangles (unchanged)Original sizeMeshopt + CleanupQuantize + CleanupBest savings
River stone (1 mat)500 tris36.1 KiB (36,944 B)6.4 KiB (-82.3%)8.1 KiB (-77.5%)-82.3% (Meshopt)
Castle exhibit (19 mats)456 tris42.7 KiB (43,732 B)20.5 KiB (-52.1%)23.1 KiB (-45.8%)-52.1% (Meshopt)
Deer exhibit (25 mats)330 tris37.2 KiB (38,044 B)25.9 KiB (-30.4%)22.8 KiB (-38.7%)-38.7% (Quantize)
Fox exhibit (15 mats)212 tris23.4 KiB (23,916 B)16.2 KiB (-30.6%)14.7 KiB (-37.2%)-37.2% (Quantize)
Forest pine (5 mats)92 tris9.5 KiB (9,716 B)7.8 KiB (-17.3%)7.5 KiB (-20.7%)-20.7% (Quantize)

Notice a practical engineering nuance: on very small multi-material props (< 400 tris), Quantize can beat Meshopt because Meshopt adds per-buffer metadata overhead; on larger meshes or textured GLBs with 2K/4K PNGs resized to 1024 px WebP, Meshopt + WebP delivers the largest reductions.

Engine Compatibility

Engine & runtime support for Meshopt, Quantize, and WebP

Choose Meshopt + WebP for modern web and Godot 4 pipelines, or switch to Quantize + PNG/JPEG when targeting legacy importers.

Godot 4.x

Quantize + WebP native; Meshopt via glTF support

Godot 4.3+ supports KHR_mesh_quantization and WebP textures cleanly; if an older 4.x build or export preset rejects EXT_meshopt_compression, switch Geometry compression to Quantize.

Three.js / React Three Fiber

Full support (Meshopt + Quantize + WebP)

Register MeshoptDecoder via loader.setMeshoptDecoder(MeshoptDecoder) for EXT_meshopt_compression. Quantize and EXT_texture_webp work in all modern browsers.

Unity (glTFast)

Full support with Meshopt package

glTFast supports KHR_mesh_quantization natively and EXT_meshopt_compression when the meshoptimizer decompression package is installed. Keep PNG/JPEG if WebP decoding is not enabled.

Unreal Engine 5 / Blender

Blender 3.6+ native; UE5 prefers Quantize/Cleanup

Blender imports Meshopt, Quantize, and WebP cleanly. For Unreal Engine 5 Interchange pipelines without extra plugins, use Quantize or Cleanup-only with PNG/JPEG textures.

常见问题

Does the GLB compressor upload my 3D assets?

No. Buffer compression, vertex deduplication, texture resizing, and WebP encoding run 100% locally in your browser Web Worker. Files up to 30 MiB never leave your device.

What are the file size and input limits for the GLB compressor?

The compressor accepts uncompressed static glTF 2.0 binary (.glb) files up to 30 MiB and 250,000 triangles or vertices. Inputs that already contain Draco/Meshopt extensions, external URIs, skins, animations, or morph targets are rejected.

Which engines and viewers support Meshopt-compressed GLBs?

Three.js (with MeshoptDecoder), Babylon.js, Blender 3.6+, Unity glTFast (with meshoptimizer package), and modern glTF 2.0 loaders support EXT_meshopt_compression. Choose Quantize if your target importer (such as Unreal Engine 5 Interchange or an older viewer) lacks Meshopt decoding.

Why does this tool use Meshopt and Quantize instead of Draco compression?

Meshopt decodes significantly faster on the web, avoids heavy Draco decoder stalls, and pairs cleanly with gzip/Brotli HTTP transfer. For very small multi-material low-poly props under 400 triangles, Quantize can even produce a smaller file than Meshopt while staying broadly compatible.

How are embedded textures resized and converted?

Embedded PNG, JPEG, and WebP textures are decoded locally, downscaled cleanly only when larger than your chosen cap (512 px, 1024 px, or 2048 px), and re-encoded as WebP, JPEG, or PNG.

What texture resolution cap should I pick for WebGL, mobile, or desktop games?

Use 512 px for small scatter props and mobile WebGL scenes, 1024 px as the balanced default for most indie low-poly props, and 2048 px only for close-up hero assets where fine surface detail fills the screen.

Should I run Polygon Reducer or Pivot & Rotation before GLB Compressor?

Yes—always compress last. Because our browser editing tools expect uncompressed static GLB inputs, run Model Inspector, Polygon Reducer, LOD Generator, or Pivot & Rotation first, then finish with GLB Compressor as your final shipping step.

Does compression reduce my triangle count?

No. Compression preserves your exact triangle count, mesh instances, and materials while shrinking binary buffers and texture bytes. Use Polygon reducer when you want fewer triangles.

Are rigged or animated GLBs supported?

No. This compressor focuses on static GLB meshes up to 30 MiB so meshes, bounds, materials, and UVs can be verified locally before export.