Compressing glTF
Compression export writes a new, smaller copy of a glTF or GLB model. It does not modify the original, and it is separate from Save, which preserves whatever compression the source file already had.
Running an export
Select one or more glTF or GLB models, then click Compress Model in the toolbar or choose Compress Model from a sidebar row's context menu.
Mesh compression
| Option | Extension used | What it does |
|---|---|---|
| None | — | Geometry is written at full precision |
| Quantized | KHR_mesh_quantization | Stores positions, normals, and texture coordinates as compact integers rather than full-precision floats. Smaller files, no decoding cost at load, and minimal effect on appearance |
| Meshopt | EXT_meshopt_compression with KHR_mesh_quantization | Optimizes and packs geometry for efficient GPU decoding. Substantial size reduction with fast loading |
| Draco | KHR_draco_mesh_compression | Google Draco geometry compression, which produces the smallest meshes and adds a decoding step when the model is loaded |
Meshopt is the default and suits most assets. Draco is worth choosing when file size matters more than load time.
Texture compression
| Option | Extension used | What it does |
|---|---|---|
| None | — | Textures are written in their original format |
| Basis / KTX2 | KHR_texture_basisu | Encodes textures as Basis Universal in a KTX2 container, which reduces both file size and the video memory the textures occupy once loaded |
| WebP | EXT_texture_webp | Encodes textures as WebP, which is substantially smaller than PNG or JPEG at comparable quality. Textures are decoded to full size in GPU memory when loaded |
Basis / KTX2 is the default and is the right choice for the web and for real-time rendering, because the saving persists in memory as well as on disk. WebP suits a consuming application that does not support KTX2.
Every combination of output format, mesh mode, and texture mode is supported, including Draco with KTX2.
Choosing settings
- For delivery to a web viewer, GLB with Meshopt and Basis / KTX2.
- For the smallest possible download where load time matters less, Draco with Basis / KTX2.
- For an asset that will be opened again in an authoring tool, Quantized or None, which avoids adding a decoding requirement.