the clouds box material

[CTZn]

I'll give more details when I'll have the chance to use a keyboard.
You will find attached an pigq with one igs linking the cloudsBoxA material.

The scattering comes with a cumulus phase function that will revert to uniform as soon as you will edit the medium (UI concerns I'm guessing). That is kind of a preview mode that works good too. HG at 0.844 is a more realistic compromise than uniform.

the shader will operate only between the min and Max altitudes, set the box geometry accordingly.

keep in mind to copycat the parameters between scattering and absorption to avoid volumes mismatches.

feedback welcome. ask questions as needed.

[Zom-B]

This is the slowest clouds I ever rendered ^^
on 1920 x 1440 I have 56k samples/sec per machine...

My result also look different then yours CTZn.
Here a example after 45min on 3 machines reaching 145spp:

[kklors]

Thanks a lot for sharing man! Nicer results than the ones I've played with from you yonosoy for sure.

Not qualified enough to tell where the performance drop comes from, but it's slower indeed.
To be honest though in both instances, the old and new shader Max Extinc. Coeff values in the range of 0.001- 0.005 render pretty fast and give enough visual feedback on the sky and are thick enough to cast shadows etc.

@Zom-B I think the result looks different because we're missing the orange bounce light from the canyons underneath located in D:blabla/nature.

Edit: Is there a reason why it's set to Bidir? I know Thomas would now tell me to stfu as uni would converge to the same result as Bidir after time, but from early lower sample tests with clouds in the 300-2k sppx range I had entirely different secondary light behavior. Would be most interested in differences in the 5-10k range as that's the maximum I will ever render in the near future.

[CTZn]

Zom-B: you are rendering using the cumulus phase function while my render is on uniform because I edited the shader after the scene was loaded. The cumulus phase function will provide the most realistic renderings of clouds but it is several times slower than with using the uniform one. The ground color bleeding must be affected accordingly.

kklors: I did observe a difference in the the earlier stages of rendering and bidir has some advantages as you stated. Pick the mode that you like the more ! GPU acc is doing well, as high supersampling values in general.

Max extinction coeff as well as the absorbtion and scattering values are at your discretion, this is a pre-release since the shader can be enhanced with your help.

As stated in the first post, edit the medium (add a blank space in one shader) and the phase function will be set to the rocket uniform phase function wich does the trick.

I'll let you guys explore this more as I'm interested in your feedback regarding the edition with parameters and what you come up with (show up, for my retribution).

[CTZn]

got any time to fiddle with the material ? honestly the scene is not optimal nor a reference, I'd expect artists to throw their own box in !

anyways, anything I could help with ? did something throw you back ? should I rather release a uniform scattering version ?

[kklors]

Finally had the chance to play around with it.
Especially dig the characteristics scale and layer amount results. Higher scales give pretty nice results for hero clouds.

Did only some quick tests with combining different sorts on different heights and put some cauchy on some as well just to see what happens.

The only thing I noticed is that the edges start refracting sort of when they intersect. Is there a way to deal with it? One could combine all meshes to avoid intersections, but that would defeat the purpose of having different cloud types on different levels.
Attached an example with Max Extinc. Coefficient 0, as the effect happens also when they are transparent.

[CTZn]

Hi kklors, thanks for your time and for the pic.

Thank you for pointing the border issue out too, since I never tried several intersecting volumes I wasn't aware of it. I am supposing that MLT is favorising this behaviour but that I'll check on myself. Convergence is key, as we know Cauchy is in question too...

[kklors]

Ah yeah not a big issue. Happens in all render modes.

One could place the intersections in a way so they won't take over too much and paint them away in post later.
If there's an easy solution to avoid it, would be good to know though : D

[CTZn]

Hi kklors, thanks for your time and for the pic.

Thank you for pointing the border issue out too, since I never tried several intersecting volumes I wasn't aware of it. I am supposing that MLT is favorising this behaviour but that I'll check on myself. Convergence is key, as we know Cauchy is in question too...

late answer but as it comes out, the cauchyB parameter is applied evenly throughout the whole volume, regardless of any shader specification.