How to reduce GLB file size for the web

Where do the bytes in a GLB file go?

A GLB holds two kinds of heavy data: textures and geometry. Knowing which one dominates tells you where to start. A .gltf holds the same data in separate files, so everything here applies to both (see glTF vs GLB).

• Textures are the images wrapped around the model: base colour, normal, roughness, metalness, occlusion and emissive maps. In a textured product model, they often make up most of the file.
• Geometry is the vertex data: positions, normals, texture coordinates and triangle indices, plus any skin and animation data. It dominates in dense scans, CAD conversions and untextured models.

File size is not the only cost. A 4096 × 4096 texture holds almost 17 million pixels and takes about 64 MB of graphics memory once decoded, however small the PNG or JPEG was on disk.

To see what you are working with, open the model in the 3D Viewer. Its Model panel shows the triangle count, vertices, meshes and file size.

How to shrink textures without ruining them

Textures are usually the quickest win. Two changes do most of the work.

• Resize them. A model shown in a viewer a few hundred pixels wide rarely needs 4096 px textures. Halving the width and height cuts the pixel count by three quarters.
• Save colour textures as WebP. Lossy WebP is typically far smaller than PNG, and often smaller than JPEG, at a similar look. glTF uses it through the EXT_texture_webp extension.

Not every texture should be treated the same. Normal, roughness, metalness and occlusion maps store data, not colours. Lossy compression bleeds values between their channels and bends the surface normals, which shows up as blotchy shading.

The 3D Compressor never encodes these data maps lossily. It keeps their bytes as they are, and only downsizes them, as PNG, when they are larger than the preset allows.

What about KTX2 textures?

KTX2 (Basis Universal) textures stay compressed in graphics memory, which helps most on phones. They need a KTX2 decoder in the viewer. The 3D Compressor keeps existing KTX2 textures as they are, but it does not create new ones.

How to compress the geometry

Quantization

By default, glTF stores every vertex position, normal and texture coordinate as a 32-bit floating-point number. Quantization stores them as 8-bit or 16-bit integers instead, through the KHR_mesh_quantization extension.

The Low and Medium presets use 14 bits for positions, which splits each mesh into about 16,000 steps along each axis. On a chair 1 m wide, a step is about 0.06 mm, so the change is usually invisible.

Draco vs Meshopt

A geometry codec goes further. It packs vertex and index data into a compact stream that the viewer decodes when the model loads. glTF has two.

Feature	Draco	Meshopt
Made by	Google	The open-source meshoptimizer project
glTF extension	KHR_draco_mesh_compression	EXT_meshopt_compression
Geometry size	Usually the smallest for static meshes	Close to Draco, especially after gzip or Brotli
Decoding	Slower, with a larger WebAssembly decoder to download	Very fast, with a small decoder
What it compresses	Mesh geometry	Geometry, plus animation and morph targets
3D Compressor preset	High	Medium
Compatibility	Needs a Draco decoder; some desktop tools cannot read it	Needs a Meshopt decoder; some desktop importers, including older Blender versions, cannot open it

In short, pick Meshopt for fast loading and for animated models. Pick Draco when the smallest possible static mesh matters most.

When should you simplify the mesh?

Simplification removes triangles while trying to keep the shape. It is the only step that changes the geometry itself, so use it with care.

• Good candidates are dense 3D scans, photogrammetry and CAD models exported at fine settings. They often carry far more triangles than the screen can show.
• Poor candidates are low-poly models, where every triangle shapes the outline, and parts that must keep exact dimensions.

The High preset simplifies meshes, removing up to half of the triangles within a small error limit. Low and Medium do not simplify the mesh.

Which compressor preset should I choose?

Every preset starts with a lossless clean-up. It removes unused data, merges exact duplicates and welds identical vertices, while keeping UV sets and empty nodes. Then each preset goes further.

Preset	Geometry	Colour textures	Normal, roughness and occlusion maps	Best for
Low	Quantized, no codec	WebP quality 90, up to 2048 px	Lossless, up to 4096 px	Close-ups and hero products where detail matters most
Medium (default)	Meshopt and quantized	WebP quality 75, up to 1024 px	Lossless, up to 2048 px	Most websites and online stores
High	Draco, with up to half the triangles removed	WebP quality 50, up to 512 px	Lossless, up to 1024 px	Thumbnails, previews, mobile pages and dense scans

How to compress a GLB or glTF file with the 3D Compressor

1. Open the 3D Compressor Go to the free MicroPlugins 3D Compressor. It runs in your browser, so there is nothing to install and no account to create.
2. Drop in your model Drag a .glb onto the page, or a .gltf together with its .bin and texture files, or the folder that holds them. OBJ, STL, FBX and other formats are converted to GLB first. You can add several files at once.
3. Pick a preset Choose Low, Medium or High. Medium is selected by default. With several files, you can give each one its own preset before you start.
4. Click Compress Compression runs on your device. Large models with big textures take the longest.
5. Check and download Click Compare to see the original and compressed model with a slider, then download the new GLB. For a batch, use Download all (.zip).

How to check the result

A smaller file is only a win if it still looks right and still opens. Before you publish, check these five things.

• The size. Each result shows the original size, the compressed GLB and the percentage saved. For an OBJ, STL or other non-glTF source, it also shows the size as a plain GLB, so you can see what the preset itself saved.
• The look. Click Compare and drag the slider across the model. Zoom in on fine text, sharp edges and shiny surfaces.
• What changed. Show what changed lists triangles, materials, textures, the largest texture size, animations, hierarchy and glTF extensions, before and after.
• Compatibility. The notes under each result name every extension a viewer needs to open the file.
• The real viewer. Open the file in the viewer or platform you will publish to, on a phone as well as a computer.

If a result comes out larger, the source was probably compressed already. Try a stronger preset, or keep the original. Once the file passes, see how to embed a 3D model on your website.

Frequently asked questions