File size goes to 70mb when exporting gltf with 50 shape keys, and fbx is only 6mb

[butaixianran]

I'm exporting Daz Character from Blender to gltf+bin files. Character is with 50 common facial and viseme morphs, without cloth, hair or animation.

File size of gltf is 2mb, but bin file goes to about 70mb.

If I export fbx from Daz or Blender with morphs, fbx file size is about 6mb.

That's a big difference between 6mb and 70mb.

And if I export gltf+bin file without shape keys, bin file is less than 6mb.

This could make gltf out of game when exporting model from blender to game engines.

Is there a solution for this?

---

Infromation:
Blender version: 3.0

Daz to Blender bridge for 3.0:
https://github.com/butaixianran/DazToBlender

[scurest]

AFAIK FBX stores shapekey positions sparsely (#1346) and doesn't store normals.

[butaixianran]

Thanks for your reply, it helps.

Now, I unchecked shape key normal when exporting, bin file is down to 31mb.

Then, we have to wait Sparse Accessors for gltf.

[donmccurdy]

Would it be possible to share the .blend and/or .glb files for this example?

If your game engine supports the EXT_meshopt_compression glTF extension I'd suggest trying to run gltfpack on the model:

gltfpack -i in.glb -o out.glb -cc

[butaixianran]

I updated a blender file also a glb file for testing:

blender file: 108mb
https://mega.nz/file/9n5W3R7I#isbVWh_xI_vnU5JpIppL9f99tOfSkFT8nW5NcrXDgA0

glb file: 75mb
https://mega.nz/file/Rjw2nTKA#boURdDt2AE6DFFFpl3vi8JDks__FuG5JdgMJoU7oknk

There is a Daz Genesis 8 free default character in it, with facial shape keys.

Also thanks for the information about gltfpack. It helps to size down glb from 50mb to 35mb.

But I can't find any information about which gltf importer support "EXT_meshopt_compression" in Unity. Most gltf importer have not been updated for years.

[donmccurdy]

Thanks! glTFast for Unity is the package I'd suggest, and does support EXT_meshopt_compression. I'll look into ways to reduce the file with sparse shape keys as a post-export optimization when I have the chance, or the gltfpack author might be open to implementing that as well.

[julienduroure]

Closing, as there is no bug here. Only space for enhancement, in case one day we manage sparse encoding.
This is not a priority for now

[donmccurdy]

As an update here — glTF Transform v3 will support writing sparse (mostly zero) data more compactly, and could be a good option for compressing files with shape keys, like this one.

Example, using a pre-release of v3:

npm install --global @gltf-transform/cli@next

gltf-transform sparse input.glb output.glb

I didn't find the original FBX to compare, but this seems to bring the size down to where you'd expect. The .bin file containing the mesh data is now only 4.4 MB.

[mostafamdy]

Hi @donmccurdy do you know how can I use gltf-transform with threejs

[donmccurdy]

@mostafamdy let's discuss this in donmccurdy/glTF-Transform#882.

[lyuma]

Am I correct in assuming this is a duplicate of:

• Encode Morph Targets with Sparse Accessors #1346

I'll put my comment there instead.