

I’d probably try to do this programatically. A collisionshape/-volume that’s calculated on-CPU and a small shader reimplementing the same shape / sine wave just visually on the GPU.
This should work if:
- the algorythm is not too complex/heavy
- you don’t want to hand-place objects
- you like working with fractals and similar math
You can also adopt the wandering clipmap solution, if the terrain will not be convex.



















That was perhaps the saddest part about leaving Reddit. I also opted for the edit-and-delete approach, because I didn’t want Reddit to use my posts to train LLMs. But man. Years of trying to be helpful, sometimes really going out of my way to address something properly; gone.
But honestly, I don’t know of any alternative. Even if I archived everything, it wouldn’t be as accessible to anyone as it used to be on Reddit.
Reddit had (and imo still has) a monopoly on searchable advice from strangers. There’s a reason they wanted Reddits data to train LLMs.