IT
Omnvert

GLB Reducer: Compress Textured 3D Models

Shrink a textured GLB to a target size, within a few percent of it. Textures are re-encoded and resized only as far as the target needs, meshes simplified with their UV seams locked, and materials, animations, skins and extensions pass through untouched.

🧊
Drop a GLB file here or click to browse
Up to 200 MB · .glb or self-contained .gltf
Choose GLB file
1
Pick a GLB
2
Set target size
3
Download
MB

The size you set is a budget, not a ladder of fixed steps: it is split between textures and triangles and each is searched to land just under it. Free savings come first (duplicate vertices welded, opaque PNG maps to near-lossless JPEG), then JPEG quality, and only then resolution and triangles, with UV seams and the edges between materials kept in place.

Pick a GLB file to continue.
🔒 Models are processed on our server in isolated memory and removed within 15 minutes of your download (at most 120 minutes if you never download).
Browser-basedRuns in your browser

This tool processes on your device; your file is not uploaded for processing.

About

A model from Meshy, Tripo, Sketchfab or Blender is rarely too heavy because of its triangles. It is heavy because of its textures: a 4096 px colour map, a normal map and a roughness map add up to 30 MB before the mesh is counted. This tool starts there. Set the size you need, for a web page, an AR viewer, a game engine or an email, and it treats that size as a budget: split between textures and triangles, each searched to land just under its share. Opaque PNG maps become near-lossless JPEG first, quality comes down next, and resolution and triangles go only as far as the target needs, so a 25 MB target gives a file just under 25 MB, not 11.

The work is done by our own native engine on the server: meshoptimizer for simplification, with normals and UVs weighed in, mozjpeg for JPEG and gamma-correct resampling for colour maps, with normal maps renormalised after resizing. UV seams stay locked and the edges where two materials meet are kept, so the texture lands where it did and nothing cracks. The original glTF is kept as it was and only the changed data is swapped in: materials, PBR settings, animations, skins, morph targets and extensions (VRM included) pass through untouched. Draco and KTX2 inputs are the one thing not accepted yet; export those uncompressed.

Files are processed on our server in isolated memory and deleted within 15 minutes of your download, at most 120 minutes if you never download. Large files go up in chunks, so a 200 MB model uploads reliably. The result loads in a viewer right on the page, with the original beside it under the same lighting, so you can judge the change before you download.

The report lists file size, triangles, vertices, texture count and largest texture, texture data, and the materials, animations and skins kept, before and after, plus the exact settings the search landed on. One file up to 30 MB a day is free; a credit pack, a 24-hour pass or a subscription covers the rest, shared with the STL tools.

How it works

  1. 1Open GLB Reducer: Compress Textured 3D Models and choose your file or enter the required input.
  2. 2Adjust the settings and preview the result in your browser.
  3. 3Run the tool; the data is processed on your device.
  4. 4Download the output or copy the result when it is ready.

FAQ

›Will the textures still line up after simplification?
Yes. UV seams and mesh borders are locked during simplification, so vertices on a seam stay where they are and the texture keeps mapping onto the surface. Simplification removes triangles inside patches, where the texture has room to stretch a little.
›What happens to animations and rigs?
They pass through unchanged. Skinned meshes keep their joints and weights, animation clips and morph targets are copied as they are. The character is the same, only lighter.
›Why do textures shrink before triangles?
Because they weigh more and their loss shows less. The budget is split so that triangles are spent more carefully than texture pixels, and each side goes only as far as the target needs. You can pin either one.
›Which files are accepted?
GLB, and glTF files that carry their buffers and images inline. A glTF that points to separate .bin and image files cannot be read from a single upload: export a GLB instead. Draco and KTX2 compression are not supported yet.
›Is the model uploaded?
Yes, to our server, where a native engine reduces it far more precisely than a browser tab can. It is processed in isolated memory and deleted within 15 minutes of your download, at most 120 minutes if you never download. Nothing is kept or shared.
›How close to the target does it land?
Usually within 1 to 2 percent under it. If the target is out of reach even at the smallest settings, you get the smallest version and the report says so.
›Does it help for 3D printing?
No. Printing needs geometry only; use the STL Compressor, which reduces an STL to a size target and repairs it on the way.