I am trying to find an even thinner way to do instancing than the ThinInstance approach already available.
The thin instance example on the Babylon JS youtube channel (Fun With Legos) shows creating an array of meshes to form an ocean made of lego bricks. In this example, the translate matrix is supplied in javascript code and then sent over as a matrix buffer to the GPU. But then the material editor is able to override the values and set them directly (happens at 22:51 in video).
My question: Is it possible to avoid setting the matrix in JS, then do all of the translating in a material, generating the positions entirely with the NME? Thus saving on the data required to be transferred to the GPU. It’s not a lot of data with a small number of instances, but with millions of instances it adds up quick.
If this is way out of scope of NME, that’s fine, but I’m curious to hear from folks with more experience with this sort of thing