Simplify meshes and encode with Meshopt or Draco, smaller GLB for web and engines.
Please wait
Compress GLB geometry only, textures stay untouched. Mesh simplification reduces face count; Meshopt (default, recommended) or Draco further shrinks geometry buffers. They are mutually exclusive. Output is an optimized GLB. Everything runs locally in your browser.
Default50%
Range0% – 100% · step 5%
Reduces the number of faces and vertices using mesh simplification. Default medium strength; increasing strength shrinks file size further but may lose surface detail. Only affects geometric density, not the container format.
DefaultOn
Applies EXT_meshopt_compression to glTF/GLB, compressing geometry buffers without changing face count. Enabled by default for runtime-friendly loading. Mutually exclusive with Draco. Not used for non-glTF outputs like STL, OBJ, or FBX.
Default. Enables Meshopt container compression.
Do not write Meshopt extension.
DefaultHigh
Controls Meshopt encoding strategy. Default 'High' generally yields smaller size and enables attribute filters, suitable for most web scenarios. 'Medium' applies more uniform quantization across attributes, which may be preferred by conservative toolchains.
Default. More aggressive compression with filters, usually smaller.
More uniform quantization; compression ratio is usually slightly lower.
DefaultOff
Applies KHR_draco_mesh_compression to glTF/GLB. Often yields smaller files than Meshopt but with higher decoding overhead. Disabled by default; mutually exclusive with Meshopt. Not used for non-glTF outputs.
Default. Do not write Draco extension.
Enable Draco (disables Meshopt).
Default6
Range1 – 10
Draco encode/decode speed trade-off, range 1, 10. Higher values favor smaller size at the cost of encode speed. Default 6 is suitable for most models.
DefaultEdgebreaker
Draco topology encoding method. Default Edgebreaker typically gives better compression for triangle meshes. Sequential can be more suitable for ordered strips or point sequences.
Default. Best for common triangle meshes.
Suitable for ordered strips and special topologies.
Shown under Advanced settings in the UI. Vertex attribute precision when Meshopt or Draco writes glTF. Defaults: position 14, normal 10, UV 12. Lower bits shrink files more but raise crack, shading, or UV artifact risk.
Default14
Range8 – 16
Quantization bits for POSITION attribute (8, 16). Default 14 balances file size and cracking risk. Too low may cause gaps or jitter.
Default10
Range8 – 16
Quantization bits for NORMAL attribute (8, 16). Default 10. Too low may cause shading artifacts.
Default12
Range8 – 16
Quantization bits for TEXCOORD attribute (8, 16). Default 12. Too low may cause texture crawling or jitter.
Drag a GLB file onto the upload area
Set simplification; enable Meshopt or Draco (not both)
Preview, then download the optimized GLB
Yes, animation data is preserved: only the geometry buffers are re-encoded. Be aware that very aggressive simplification can deform skinned meshes, so check the preview on animated poses before downloading.
Meshopt is the default: it decodes fast on web and mobile, and Three.js supports it well. Draco can produce smaller files, but usually costs more decode time. The two options are mutually exclusive, so enable only one.
Yes, the result is an optimized single-file GLB with the same name and structure, so existing loaders and pipelines keep working. GLTF input is accepted as well and is also exported as GLB.
Yes. There is also a dedicated GLTF vertex compression page; both accept GLTF and produce an optimized GLB. Upload the .gltf together with its .bin and texture files so nothing is missing.
Reductions of 50-90% are typical, depending on mesh complexity and settings. Dense models with high polygon counts benefit most from Draco or Meshopt, and simplification adds extra savings by removing faces first.
Yes. The real-time preview shows the simplified mesh, so you can compare detail loss and animation behavior. Adjust the strength until the silhouette and motion look right, then download.
How many bytes does vertex data actually take? Quantization, MeshOpt, and Draco - which tool fits which scenario? It's not about 'Draco is best,' but 'the right tool for the right job.'
Take a 50MB PBR model through the full pipeline—Blender export, vertex compression, KTX2 texture conversion, engine loading—and get it under 5MB. Includes a complete automation script and a series cheat sheet.
ETC1S or UASTC? How to use toktx and gltf-transform? How to load KTX2 in Three.js / Babylon.js? This article is full of copy-paste-ready commands and code.
Desktop, mobile, VR, mini-programs—which compression combination should you use? This article consolidates the knowledge from the previous four articles into a decision matrix and an automation script—just follow it.
