Yo @Deltakosh or @sebavan … Is there some kind of WEBGPU compatibility mode or something like that for WEBGPU on the playground that i am missing when trying to use in a boiler plate template ?
I can load the samplescene.gltf and playerarmeture.gltf just fine on the playground in webgpu:
But when i create a app from scratch or use the vite boilerplate template (or any react project)
I get this:
Note: WebGL2 runs fine. Its just webgpu in a project outside of the playground
Looks like your code use too many varyings 
limited to 16 at the moment
by default
Not sure if your device can bump it but in the webgpuEngine ctor there is an option to request the max limits 
But it runs in the playground just fine:
I am not using any custom shaders
Yo are da man @sebavan …
its actually called setMaximumLimits=true on the WebGPU options
I got another one for @sebavan
For some reason on ES6, after 8.0 (may > 7.50) my same model that works fine UMD and did before, now kinda SQUASHES down
I have no idea what it could be 
@bghgary any clue ?
It looks like meshes are the same size, but connection points are offset vertically during movement. Maybe something to do with bone or animation definitions?
And it only happens on ESM after V8
Hmm, I don’t see how UMD and ES6 can change the size of a model.
Me neither… But the UMD version, loaded here on the playground works just fine:
Yo @bghgary and @sebavan @Deltakosh
I made two repos…One UMD and one ESM… same App.tsx and Viewer.tsx (excluding the imports)
The ESM version squashes when you run (hold shift to run) and land from a jump
Check you please check them out, see what you can find:
npm install
npm run dev
Thank you so much
Can you try to simplify the code ? This is way too much to troubleshoot efficiently. Can you repro without the unity plugin and only the main character ?
Im afraid not, its a unity exported player model.
Its really weird, that ES version is all squashed and the UMD is fine. Is same code for both.
@bghgary … Just checking, have you made any skeleton type changes in the GLTF parsing since V8 ?
I don’t believe so.