Clustering at runtime

Looking for a way to generate real time clusters similar to how HLOD would work.

The issue is that in a modular environment the HLOD system creates pre-baked meshes.

My theory so far would be to update the default material of the HLOD to masked, and to actually procedurally mask out the objects that go missing to prevent obvious popping.
Problem is, I have no idea how to determine what UV needs to be masked out. Since generating hlod creates its own random UV set for the merged models.

I’m wondering if maybe there is a better runtime way that I have overlooked which would lead to draw call aggregation.

If you will,
Imagine an asteroid cluster where there’s billions of asteroid, but as you get closer to them the models go from being impostor groups to individual asteroids.
At any time, one of the asteroids can go “missing” - and this change has to be mirrored down to the HLOD.
(Easiest example I can think of right now).