Model-viewer: Ability to position model external to glb/gltf content

[gjcope]

It would be helpful to be able to translate and rotate content external to the model file so that it loads in a desired position. Perhaps through additional tag attributes.

A good use case for this is when loading content from a 3rd party that you don't have edit control over but want to load in a specific orientation.

[cdata]

This has come up in other discussions we have had. I think a top-level rigid transform is not out of the question for us, and has been demonstrated to be desirable for some use cases.

That said, a lot of cases can be handled by the attributes we already have: camera-orbit and camera-target. These attributes let you position the camera relative to the model, which will have the same end result in many cases. Have you tried using those attributes?

[gjcope]

Yes, camera positioning will definitely work in some cases but a common use case for us (Smithsonian) would be more about relative positioning of objects from different source files (which I know is also not yet supported) which would need rigid transforms.

[cdata]

@gjcope thanks for confirmation!

For the multi-object use case, the only current idea we have floated is to enable that via 3DOM, peusdocoded:

<model-viewer src="some.glb">
  <script type="experimental-scene-graph-worklet" allow="node-hierarchy; node-transforms; load-models">
addEventListener('model-change', () => {
  // Translate the model loaded from 'some.glb'
  model.scene.nodes[0].setTranslation(...);


  loadModel('other.glb').then((otherModel) => {
    // Translate the model loaded from 'other.glb'
    otherModel.scene.nodes[0].setTranslation(...);
    // Append the other model to some model's scene graph
    model.scene.appendModel(otherModel);
  });
});
  </script>
</model-viewer>

Although this theoretically gives you the capability to do what you describe, I wonder if there is a corresponding declarative form to accomplish something similar (e.g., put one model inside another, and apply transformations to both).

At any rate, I think a top-level rigid transform may make sense for us to add.

cc @elalish for additional commentary.

[stevesan]

A top-level transform attribute sounds like a good idea. Even for the single-object case, you want it to be oriented properly relative to the world. Otherwise, for a model that is flipped on its side, camera controls would get confusing.