Eevee/Workbench development updates

Made a quick test scene for that new dof and I gotta say that Eevee dof is finally usable. Great job Clement!

Before:

After:

14 Likes

Damn ! It’s looking so clean ! Definitely on par with Cycles’ DOF, as far as I can tell.

New DoF landed on master branch, so from tonight it will be available in buildbot 2.93 alpha build.

14 Likes

Workbench: Improve AntiAliasing sampling, by Clément Foucault (task, not in master yet)
https://developer.blender.org/D10414

Theses improvement may not be very useful for every day case but it was an
experiment to try to make TAA usable for GPencil.

1 Like

pls, take a look at pointlessness so we can have more of a completely procedural workflow between Eevee/Cycles and ditch uv mapping.

is vulcan aiming for blender 2.93?

Not sure there is an ETA on that, it’s worked on but probably it will be in several pass, don’t hold your breath …

1 Like

Already in master some fixes and improvements for EEVEE Ambient Occlusion (Refactor), by Clément Foucault:
https://developer.blender.org/rB64d96f68d6ef0411383bb46d1a95d47769d927e5

9 Likes

oh nice! i’ll be interested to see what these improvements look like, eevee’s AO already felt pretty decent.

So I tried it out now and tbh I don’t like the new one at all, looks quite a lot worse than the old one.

This is from the latest 2.93 alpha
ao_eevee_293

2.91
ao_eevee_291

And one from Cycles as a reference
ao_cycles

As you can see the new one looses quite a lot of that nice contact shading, for example between the tires and fenders, the tires and the ground, the number plate and the lid onto of the front. Sure the new one have more correct brightness compared to the reference, but that can easily be adjusted in the old one if you want to.

So for me this is clearly an downgrade, unless the latest alpha have some kind of bug that is?

1 Like

If you are able to share a .blend file with developers where it is possible for them to reproduce what you think is a downgrade, you can report it on:
https://developer.blender.org/

I was actually thinking about this, but wasn’t sure if this would be considered a bug. But I might do it then, thanks for the tip :slight_smile:

Bug tracker is open to Artists. Artists are not required to know technical details. Artists simply observe something that may be wrong and report the problem, and then developers are the ones who decide whether or not the problem is considered a bug.
So you report without fear, and explain as best you can the disadvantages that you see in new AO implementation.

2 Likes

Perfect :slight_smile:

Another test, this time I do see a big advantage for 2.93. The old one do have big faults here with “shadows” that shouldn’t be there from many of the legs. Thou I it do have the nice contact shading that the new AO do miss.

2.93

2.91

Cycles

4 Likes

Clement added in a factor that reduces the shadowing of thin objects, like those spider legs.

The factor is currently hardcoded, but I bet exposing it to users would help a fair bit.

Overall the new AO does seem to be an improvement, but it seems to be much more subtle than it was before.

1 Like

2.91 AO

2.91 Render EEVEE

2.93 AO

2.93 Render EEVEE

4 Likes

For those getting more subtle results with new AO. Keep in mind that “factor” value can go higher than 1 if you enter the value manually. “1” should be a safe value, but this does not prevent you from entering the values manually. Have you tried this?

Anyway, for those getting unwanted results with new AO, it would be better to discuss it with developers by sharing a .blend file on the bug tracker or developers forum:

Even if you think that default values should allow stronger results.

You do not consider the glossy problem there, there is a problem with it on 2.93:
https://developer.blender.org/T85908

Some considerations in that report about the new AO by Clement:

This is caused by 3 things:

  • The AO is more correct as it gives less dark values on some regions.
  • The number of samples was considerably reduced leading to less features for really high radii. This is planned to be fixed.
  • The thickness heuristic does not yet have a parameter to tweak it. This might reduce shadowing. This is planned to be fixed.
2 Likes

The AO clearly improved on places like floors with legs or other thin elements.

I guess hair/fur falls under reduced “shadowing of thin objects”.
In the attached image the AO factor was cranked up in 2.93 (as seen on the tongue), but the hair remains very subtle/bright. It would be great if this can be controlled, to return some AO definition in hair.

EDIT: here just a look at the AO pass.

7 Likes