Lumen Translucency lighting vs opaque mismatch is larger than expected

Development Rendering
, ,
1

Our VFX team has been struggling with getting their Niagara systems with translucent materials to fit into the environments they are being used in. In a project with default settings, I was able to recreate a test scene where an opaque material had a pretty significant lighting mismatch compared to a translucent material with an opacity of 1. I’d never really expect them to perfectly match but this difference is notable enough that I wanted to raise a ticket here. I wanted to ask:

  1. Is this much disparity expected?
  2. If so, are there any improvements in the works in this space?

As a bonus question, while looking into this I was looking at the radiance cache visualization mode and it looks like the minimum trace length specified in TraceRadianceCacheProbeTexel is causing a surprising amount of potential bleeding for the radiance cache samples. The two screenshots below demonstrate the core issue I raised above and also show the radiance probe samples with

Ray.TMin = max(max(MinTraceDistance, RadianceCache.ProbeTMin), TraceHitDistance - PullbackBias);

vs a very hacky test value of this

Ray.TMin = max(0, TraceHitDistance - PullbackBias);

Making this change doesn’t actually have an impact on the core problem I’m trying to solve but I have confirmed that the translucent object is getting it’s GI from the radiance cache so I imagine this will have an impact eventually.

2

Steps to Reproduce

  1. Create a solid material and a translucent material with opacity of 1.
  2. Create a lighting environment that exaggerates lighting from Lumen
  3. Apply the 2 materials to 2 spheres in the environment
  4. Observe the lighting differences between the two being fairly significant.
3

Hi Joe,

Thanks for reaching out. I am still waiting for feedback from the developers about whether this is a design issue. I will get back to you once I have some more updates. Thanks for your patience in the meantime.

Cheers,

Tim

4

Hi Joe,

Sorry for the long radio silence, but I have an update for you: The discrepancy in the amount of lighting the translucent sphere has is more or less by design. Translucent materials receive their lighting from the world space radiance cache, which you already pointed out is at a lower resolution. The screen probes also don’t trace the center of the probe, so at times, you might be picking up some light leaking. I could not confirm this in the sample project, but does moving the translucent sphere away from the surface affect the lighting? If that is not the case, I will file a Jira ticket with the developers so they can take a closer look. Let me know what you think.

Cheers,

Tim

5

When I move the spheres, I think there is some amount of AO showing on the translucent sphere when it moves closer to the wall but I don’t see any hints of red on it. It remains pretty desaturated and I would expect some amount of the saturated red to be showing via the SH based radiance cache.[Image Removed]

6

All of that is to say that I do think it’s worth raising a ticket with the developers for this.

Thanks!

7

Hi Joe,

No problem at all. I have created a ticket with our developer team to investigate this issue further. You can follow the progress of the ticket here: https://issues.unrealengine.com/issue/UE-318661. Please let me know if there is anything else I can do for you.