Compression

Geometry and quantization

Mesh cleanup and vertex precision

Geometry settings change the mesh data inside the GLB. Some changes are mostly cleanup. Quantization is different: it stores vertex data with less precision, which can make files smaller and can also change how the model looks.

Geometry cleanup

Open Advanced > Geometry to control mesh cleanup.

Geometry compression settings with Weld Vertices, Deduplicate accessors and textures, and Reorder for size enabled

OptionWhat it meansTurn it off when
Weld VerticesMerges identical vertices so shared geometry can be stored more efficiently.The model shows unexpected seams or topology changes after compression.
Deduplicate accessors/texturesRemoves duplicate GLB data such as repeated accessors, textures, meshes, and materials.Rarely. This is usually safe.
Reorder for sizeReorders mesh data for better binary compression and transmission size.Another app has trouble loading the compressed GLB.

Unused data is pruned during compression even when no visible setting refers to it.

Quantization

Open Advanced > Quantization to control how precisely vertex attributes are stored.

Quantization settings for Position, Normal, Texcoord, Color, and Generic attributes

Lower numbers usually make the file smaller. Higher numbers preserve more precision.

That tradeoff matters most for CAD-derived models. A tiny hole or sharp edge can be useful geometry, not decorative detail.

OptionRangeDefaultControlsIf too low
Position8-1614Vertex positions.Edges, holes, or small CAD details may shift.
Normal6-128Surface normals used for lighting.Surfaces may look faceted, wavy, or shaded incorrectly.
Texcoord8-1612UV coordinates used to place textures.Textures may smear or no longer line up with the model.
Color6-128Vertex colors.Gradients or painted colors may show banding.
Generic8-1612Other vertex attributes not covered above.Custom data or tool-specific attributes may lose precision.

Common fixes

Edges moved or small details disappeared

Increase Position. For CAD parts, use Low or set Position to 16.

The model looks faceted

Increase Normal. If the source file already had rough normals, convert again from the source and inspect the uncompressed GLB before compressing.

Texture placement changed

Increase Texcoord. If the texture was already wrong before compression, check the original conversion and sidecar files.

Another app rejects the GLB

Use Low first. If it still fails, disable Reorder for size and keep higher quantization values.