Interesting, never noticed any difference between those two modes before, so I thought it was just a fallback when not using MLs. Will check tomorrow.
BTW they have “updated” the project to 5.7 today. If it were deprecated, they usually stop updating them. Anyway I think it’s an issue mainly related (but not only) to grooming.
Unexpected results, from my POV:
ML RT (artifacts):
ML VSM (clean, but maybe missing grooming shadowing):
No ML, pure RT (fine):
PS: I would have thought the artifacts would happen in the VSM method, not in the RT one.
It seems like MegaLights is somewhat cheating as the quality of shadows is significantly degraded. Screenspace shadows are almost completely gone with megaLights enabled.
MegaLights On
MegaLights Off
Just like others have already reported, 5.7 has some visual glitches that were not present in 5.6 such as screen traces casting shadows in places where they shouldn’t exist.
The first image showcases the problematic shadow with r.MegaLights.ScreenTraces=1, the second image uses r.MegaLights.ScreenTraces=0 which fixed the issue.
Also, while not directly related to this visual issue, 5.7 does appear to improve overall shadow responsiveness. However, compared to 5.6 it is noticeably noisier, particularly in specular highlights.
I actually wanted to ask the Megalights team about that: are there any plans to improve the Megalights forward translucent implementation? The quality difference isn’t bad on small objects, but things look pretty shaky on large or curved pieces of glass. In areas with multiple local lights, things look pretty weird.
There shouldn’t be anything preventing MegaLights specifically from working on Vulkan. Just tested this in 5.7 and seems to be working as expected.
Can you share more information / logs when you run using Vulkan?
There seems to be a bug in the logic that estimates the surface thickness for screen traces. In 5.7 we extended this logic to increase the thickness of surfaces that are not represented in HWRT, however there seems to be some regression which is making all objects incorrectly use the increased thickness. I will investigate and work on a fix.
In the mean time you can try reducing r.MegaLights.ScreenTraces.RelativeDepthThicknessWhenNoFallback to have the same value as r.MegaLights.ScreenTraces.RelativeDepthThickness.
Yes, there’s ongoing work to improve lighting on translucent materials in a future release.
Yeah, currently specular on forward shaded surfaces uses a very simple approximation which doesn’t work well when there are multiple competing specular highlights.
In my case that helped with the strange shadows on the Meerkat example I posted.
The strange shadow gone when changing r.MegaLights.ScreenTraces.RelativeDepthThicknessWhenNoFallback 0.01 to r.MegaLights.ScreenTraces.RelativeDepthThicknessWhenNoFallback 0.001
Not sure if it will have a noticeable tradeoff somewhere.
The combination of Megalights with Instanced Stereo (for VR) gives visual artefacts near dynamic lights. Either one or the other has to be switched off (UE5.7).
Steps to replicate: start from 3rd person template, add VR content package, turn on megalight and Instanced stereo and view in VR.
