What is causing these adaptive subdivision pop ins?

Hi everyone,

So I am working on a few projects at the moment that involve zooming in to an extremely small point. I am getting close but I keep running into problems with adaptive subdivision.

There always seem to be moments when the subdivision pops into the next LOD level. I’m doing a render now to see if increasing the dicing rate to .25px will fix this but I would love there to be another solution because this adds insane render times… But my understanding is that it should be updating every single frame so why would there be clear lines?

This is how I see it behaving right now.

This is how I would like it to be.

Here are some tests where you can see it happening.

In this one please ignore the sand particles. I have figured out a way to make those transition in a much smoother way that makes the transition less obvious. But you can see before the sand spawns in there are very visible lines where the LOD is changing. The effect is really visible at 8-12 seconds before the sand comes in over it.

You can see it happen in this example too. It’s less obvious but I think that is because the camera is not travelling perpendicular to a plane. This is the one I am rendering out with a .25px dicing scale. I really hope there is a better solution to this than just cranking this value up really high.

If anyone has any ideas. I would appreciate your input.

I haven’t used LOD myself yet, but as far as i know the object are choosen from different LOD version of one object (maybe procedural produced). And because of this some object switch details.
Anyway: Is there a way to produce two images per frame where for example one specific object is in LOD 3 and in the other in LOD4 say for frame 42 to 49 and for the endresult do a composite fade betwen them?
Maybe this will reduce detailswitching. The 3d renderwork will be doubled and a bit more performance would be used to make the blend, but better than rendering everthing in full detail.

This is more for the adaptive subdivision problem. Object LOD I have figured out a somewhat solution with geometry nodes.

But the adaptive subdivision LOD is one single mesh and it subdivides the most when it’s closes to the camera but it’s producing pop ins when I don’t think it should be.

Ahh, geometry nodes… i have just recently looked into them (after anime node, sveroc? ) That wil be fun :smile_cat:.