Page MenuHome

Marbling effect on Eevee and subsurface scattering...
Closed, ResolvedPublicBUG


System Information
Operating system: Windows-10-10.0.18362 64 Bits
Graphics card: GeForce GTX 1070 Ti/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 431.60

Blender Version
Broken: version: 2.81 (sub 12), branch: master, commit date: 2019-09-25 13:57, hash: rBf8da7218be13

Short description of error
Eevee backlight causes marbling light effects...with subsurface scattering

Exact steps for others to reproduce the error
Created Suzanne model, added backlight. (included screen shot and test file)

Event Timeline

I tested the version in 9-13-2019 as well and the marbling is also there. Not sure about earlier september versions.
I noticed also that changing the 'count' in the cascaded shadow map changes the marbling effect. (video capture included):

Philipp Oeser (lichtwerk) lowered the priority of this task from 90 to 80.

This is due to rBd8aaf25c23fa: Eevee: Shadow map refactor.

There are some things mentioned in D5659: Eevee: Shadow map refactor regarding SSS..
There is also more information here
That all being said, just opening your file in 2.81 after the shadow refactor will not give you the same results.

However, by tweaking some settings, you should be able to get comparable (if not better) results.

  • I think the main point is to enable Soft Shadows under Properties Editor > Render Properties > Shadows (this was previously a per light setting)
  • upping viewport samples Properties Editor > Render Properties > Sampling also helps

Dont think this counts as a bug, but will let @Clément Foucault (fclem) close...

Bug or not, this is an undesirable behavior that makes SSS quality awful, to the point of unusable, this needs fix.

@Jean Da Costa (jeacom256) : have you tried enabling Soft Shadows and setting Viewport Samples to 64 upwards?
(this seems to have no marble artifacts for me -- can you show where this is still awful?)

I agree though that translucency looks different from before [distribution seems to be different, cannot be as soft as before?], maybe @Clément Foucault (fclem) can comment on that...

All those things may make it less noticeable but it doesn't solve it. More samples don't make it converge either.
Here's an example with with big radius, higher res shadows, soft shadows, lots of sss samples and viewport/render samples, high bitdepth, etc. Still visible

Crucially, if you turn off shadows for a particular light, SSS translucency will disappear.

There's at least one improvement. With VSM shadows in the past, the middle of things would get darker, like the ear. Now it's closer to Cycles. In any case, I think the line needs fixing, otherwise translucency is almost useless

@Philipp Oeser (lichtwerk) enabling soft shadows or upping samples only seems to smear out but not remove this weird line.

2048 samples, soft shadows enabled and the problem is still there:

Thx for getting back now I actually get it better what is meant by 'marbling'.
Tbh, I am not sure what the origins of these are [or how to get rid of them], I think @Clément Foucault (fclem) has to chime in here...

I am convinced that it was caused by D5659 might be some biasing issue.

Clément Foucault (fclem) lowered the priority of this task from 80 to 50.Sep 28 2019, 12:58 AM

It is caused by my refactor yes.

This is a side effect of having biased shadow maps. Before they were smoothed but I remove the smoothness properties.

A better ways would be to introduce a corrective bias to make it match perfectly the depth that is supposed to be in the shadow map.

Clément Foucault (fclem) raised the priority of this task from 50 to High.Oct 15 2019, 3:06 PM