Looking more into SVOGI, when stepping through its logic with a debugger, it does seem to go though meshes and tries to voxelize them, I think I see why expected effect is not visible though.
I set SVOGI to integration mode 2, which would reflect gathered voxels in mirror like materials, turned off screen space reflections and setup a bunch of mirrors throughout the level. Also had to set hdr max to very dark, otherwise reflected voxels are too bright, but that is why example below is so dark.
You can see below that part of the red sphere is being revoxelized as I move it, but
- Only parts of it(It doesn’t voxelize more with time).
- The mirror on the right is away and quite a bit above the sphere, so it should not be reflecting it at that position, so its like voxels are positioned somewhere away from their originating meshes.
- When spehere is moved along x(red) axis, it moves in the opposite direction in the reflection, so maybe thats part of the issue, x axis is somehow flipped for voxel vs meshes?
- The voxel bricks(yellow) are not square, they are squished for some reason, that seem to also contribute to the incorrect voxelization.
From other posts It sounds like after LY 1.14, SVOGI was moved to its own GEM and kind of got broken in the process, there are few posts, but here is a good example:
How can we get the same effects in 1.16 with 1.14 by setting SVOGI parameters? - Lumberyard Discussion / Art & Graphics - Amazon Lumberyard Game Dev Community (awsgametech.com)
Also I see some recommend to increase shadow bias and SSDO for better results. As I understand, shadow bias simply makes shadows less dark, and SSDO amount property there to reduce non GI AO, not to increase it, since SVOTI is suppose to bring more of its own AO, ssdo can be increased without svoti on.
I tried getting 1.14 from GitHub to compare visual results, but supporting files for git_bootstrap.exe are no longer available for that version, so I’m unable to compile/build it.
In any case, I would love to see maybe a separate website specifically for features and GEMs for LY, where feature requests can be submitted, upvoted/downvoted, also existing features and GEM’s can be rated, or show/let submit current issues that also can be rated by importance to users. I would think that would help LY team to determine what is in demand or currently is not working for many users, helping to shape the roadmap in a flexible way.