[eric nixon]

Hi Ben, I was wondering how you dealt with the leaves and clipmaps, because the last time I rendered a tree, the intersection of many leaves which were clipmapped caused a great slowdown. Did you just avoid clipmapping? or avoid intersecting leaves?

[Tora_2097]

Hi Eric,
I avoid clipmaps whenenver possible. Especially for trees in the distance or when using motionblur I disable clipmaps alltogether. That said it is not entirely feasible to dispense with the clipmaps because of the otherwise enormous polycount. Still I do not find them to be that slow .
The leaves on the trees are intersecting -currently there is no way for me to avoid that- and it has an impact on rendertime but it was only for the cherry-trees where there are tons of tiny intersecting clipmaped blossoms in the V-Ray version where I noticed a huge slowdown in Maxwell. I updated the blossom parts then to not use clipmaps and averaged the shape by using more polygons. On the normap trees it renders just fine benching around 200 if I isolate a single tree.
The BM for the various scenes here are between 40-90 on an i7 980X stock speed, the ones with motion blur being the slowest. Some scenes rendered in 30m some in 2.5h.
So in general it is a good idea to avoid clipmaps but they work decent enough on normal trees. Its actually the same in V-Ray or any raytracer.

Ben

[eric nixon]

Thanks for the detailed info, Meshlab has a 'select intersecting faces' function, but this is only helpful if the majority of the leaves are not intersecting, otherwise it selects too many leaves.

Fingers crossed for some optimization in this area, + a double sided thin-sss material, and also a working thin-sss re. emitters and ofcourse the colour inversion issue with thin-sss.... hmmm so many reasons to wait before making a better tree, I think trying to make virtual nature will always deliver quite limited realism, but practically we do need to offer clients some solution, I still use 2d trees whenever I can..

[Tora_2097]

Thats true competing with nature is a daunting task since as an individual you have very limited resources and are confronted with a seemingly unlimited amound of data. Each and every leaf is twisted and bent in a unique way, bark can show the weirdest displaced surfaces and variations. I will also use 2D trees when appropriate, but most often I am confronted with animations and it quickly falls apart then.
Anyway, here are 2 more examples:

[Hervé]

very cool Ben.. the second one looks like a painting

[eric nixon]

Its beautiful

[eric nixon]

This leaf mxm is very fast to render, 10min sl17, I would recommend this method, just a diffuse bsdf and one thin-sss bsdf, maps and original mxm by bubba. check your pm perhaps its useful.

[rusteberg]

Ben, they look great except the last two look a little flat to me... (those last two in particular look better if I squint at them..... what happens if you take those two mxi's and dial down the gamma to 1.8 - 2.0?)

[gadzooks]

Eric is that mxm of Brian's leaf for sale on his turbo squid account. Or did you tweak it to your liking. I ask because it looks really really good! Mind showing you BSDF setup for that leaf?

[Tora_2097]

Thanks everyone!
@rusteberg: Did that already. In the daylight version its probably the angle of the sun causing a flat look.

@eric: Well thanks for your recommendation and your efforts with the MXM. As it happens it renders a little bit slower than my 4 layer approach. I have a mostly diffuse base + Spec layer+ SSS Layer+ Ghost layer to clip through all layers underneath. That way you can quickly dial up/down, turn on/off any component or disable clipmapping altogether. I noticed a much stronger scattering in your version but a very weak specular component. I guess its rendering slower because of the scattering contributing more the result.

Ben

[eric nixon]

@gadzooks, I was hoping Ben might have time to render a tree test with such a leaf setup, if he does I'll upload the mxm the tweak was only, to put everything in one layer.

[numerobis]

it's really funny that we have ThinSSS for almost 4 years now and there is still no guideline from the inventor how to optimize
the settings for various cases to get the fastest and best results and that there are only some "knowing" people who keep their settings a secret...

(btw. a big THANKS to rusteberg for all these tests! )

[eric nixon]

I think you should thank Ben actually.

Numerobis, maybe the fact that thin-sss renders wrong under emitters is a secret?

So, with that info it would be wise to avoid preview scenes in mxed which use emitters.

Other than that, thin-sss bdsf should have very dark grey R0 or black to get the most sss effect from that bsdf, put all the diffuse in another bsdf, that diffuse can have maximum brightness. Put both bsdf in 1 layer.

Maximum scattering occurs around nd 1.5 assym -.5 in my experience, so thats a good starting point. To max the scatter - increase the scattering coeff until you see artifacts then back off to a safe number, often 1500 or 2000 works with a maxed out colour chip.

I totally agree there should be more documentation about this after this time, I suspect the emitter issue needs to be resolved before an official tutorial is released.

[numerobis]

Thank You Eric for the info!
It confirms some of my own discoverings, but some points are new...
I had a 3 layer approach lately (sss, diffuse, spec in normal blending mode) because i always got some dark black areas in my trees using only one layer.
But i have tested it again and it renders almost twice as fast when i have the tree BSDFs in one layer!
I have to test now where the darkening comes from. Maybe it's only an overall higher contrast or intersecting faces?
If i remember correctly the last time i tested with one layer, the leaves tend to render black at grazing angles when the ref 90 colour in
the sss-BSDF was not black.
Concerning the scattering coeff. i recently worked with very low values of 1.0 or 0.1 to get more contrast between leaf colour and
scattered light, but maybe that is a wrong assumption...

The problem with emitters is the inverted colour? Or what do you mean?

[Tora_2097]

without a cause: