Page MenuHome

Cycles: Default changes for Blender 2.77
Closed, ResolvedPublic


here a list of defaults I would change for Blender 2.77.

Done (already commited):

Render settings

  • Samples (final render): Would increase from 10 to 50 or 100. Especially beginners often wonder why their image is super noisy, and you barely ever stick to the 10 samples anyway.
  • Filter: Change from Gaussian to Blackmann-Harris, it's a bit better as seen in Example images in D1453. Not sure we have to adjust the default Width then?
  • Tile Order: Change default to Hilbert Spiral. This one is slightly faster than the current Center method.

Material settings

  • Volume Sampling: Default to "Multiple Importance", rather than "Distance". The best of both worlds.

Lamp settings

  • Enable MIS for lamps by default. This one might be a bit controversial, but I have seen people who missed Caustics / light through Glass and couldn't figure out why they are missing.

Suggested changes:
World settings

  • World MIS: Enable this per default. We can auto disable the feature internally, if we don't have an HDR or a Texture attached to the Background node, so there will be no performance loss for simple scenes. This option could even be removed in the future, but that's for another time.
  • World MIS samples: Decrease to 1, would be consistent with Lamp samples, and the world is just another light type.
  • World Volume sampling: Also change to Multiple Importance? (Defaults to Equiangular atm)
  • Ambient Occlusion Distance: Needs more discussion.

Render settings

  • Make spatial splits the default (needs more work first, Sergey will check).



Event Timeline

Thomas Dinges (dingto) raised the priority of this task from to Normal.
Thomas Dinges (dingto) updated the task description. (Show Details)
Thomas Dinges (dingto) added a project: Cycles.
Thomas Dinges (dingto) set Type to Design.

I generally agree with the changes, although that auto-disabling might be a bit problematic, since there'll be no way to force World MIS in that case...

Leaving the filter width at 1.5 should be fine with Blackman-Harris.

There is no point in having MIS in cases were we have a uniform color. I would check for the case where we only have a Background Node -> Output, and disable MIS in that particular case.

For volume rendering it might also make sense to enable MIS by default.

There is no point in having MIS in cases were we have a uniform color. I would check for the case where we only have a Background Node -> Output, and disable MIS in that particular case.

Check how shader->has_heterogeneous_volume is implemented, you can do the same for detecting constant background colors. That covers a few more cases and would be more consistent.

Like Ton, I like the philosophy of Blender not to think for the user. It's really good to have better defaults and I agree with all of them, but I would let a way to make it manual.

Blackman-Harris is very good on simple scenes or on noise-free renders, but in our tests, Gaussian is still much better in mid-heavy noise images. Because there are really good denoiser like the one from altus for Pathtracing, I would leave the Gaussian and box filter available. Blackman-Harris can be the default.

Otherwise all good :)

@Brecht Van Lommel (brecht) Thanks, will check on that. Volume MIS might be a good idea too. Will add it to the list as well.

@mathieu menuet (bliblubli): This is only about new defaults, existing files won't change nor will people lose options. :)

there's one tiny thing that drives me sometimes crazy - the default for world samples (branched path) is 4 and that's way to high (especially with the square samples option ticked) it should be 1 or max 2 :)

@Thomas Dinges (dingto)

"This option could even be removed in the future, but that's for another time."

We had a project a few months back which required a nighttime HDR sphere... Enabling MIS on the world caused really long render times for not much gain with noise quality. We rendered it with a HDR sphere (+ some other node magic) with MIS disabled. Having the option to disable MIS is important.

Other Cycles based Defaults i would like to suggest are:
Changing the AO Distance to 2 instead of 10.
Having the animated seed enabled by default
Having persistent images turned on
Render Samples 100, Preview samples 50

Proposal seems fine to me.

@Thomas Dinges (dingto), let me know if you'll run into difficulties with detecting when MIS could be disabled. Proposal here tho: commit changes related on disabling MIS for constant background, and do actual defaults change separately.

I will be investigating spatial split BVH and will try to make it ready to become default as well.

Definitely agree with all your suggestions.

I can't really think of anything else, other than perhaps increasing the world MIS resolution to 1024. I generally need to put that on at least 4096 for HDRIs with strong small light sources like the sun, but I guess for most HDRIs 1024 would be enough.

Sorry for late reply. Agree with suggestions.

  • Filter Glossy settings? A sane value could get rid of LOTS of fireflies...don't know if anyone might not want a little bit of accuracy loss though. I personally use it always at 0.5-1.0
  • Indirect clamp? Really helpful, though i only get advantages when set to less than about 3.0 and might be a bit aggressive.
  • About AO distance, 2 is still pretty high IMHO, it's 2 meter if people use real world measures, I would not go above 0.5. But that depends if modeling in real world scale (which is quite recommended with Cycles).

@Marco G (marcog) I disagree with clamping, that is a bit too much imho, but Filter Glossy would be fine imho. This can avoid quite some evil fireflies.
@Brecht Van Lommel (brecht) @Sergey Sharybin (sergey) Atm we are still pretty much unbiased with our defaults, do we want to change that? A little bit would be fine I think, but not too much. What do you think?

We should stay as much unbiased a possible. There's also no real scene-independent value for clamping, so that should stay as it is now.

Filter glossy is the only place where we might violate the unbiased rule perhaps and only do this if there's no measurable difference in lighting of scenes which doesn't use it (as in, we don't explicitly shooting self foot by disabling important light details).

Maybe increasing the default size for lamps would be also usefull, as far as this is one of the main reason behind noisy renders.

Hi, sun size is a long term todo for Cycles and BI.
A CGSociety user post a video how default setting should change, he calculate 0.005.
Spoke with Brecht long time ago and he mention it is correct, more or less.
kevindietrich gave me hint to a patch change sun size to angular size, but BI implementation is not included.

My proposal is to change default sun size to 0.005.


Cheers, mib

Thomas Dinges (dingto) closed this task as Resolved.Feb 5 2016, 10:51 PM
Thomas Dinges (dingto) claimed this task.

Commited second batch of changes (World MIS):

That's it from my side for 2.77, I am not convinced about the other suggestions yet, but we can talk about it again for another release. :)

The one thing I would change still is enabling animated seed by default. It's practically always what you want for animations, to the point where I would almost suggest removing the option.