cloudproblem

Started by Dune, June 08, 2020, 08:58:12 AM

Previous topic - Next topic

Dune

Two problems I have to solve. First cloud. I wanted to have a cloud with shadows of surfaces turned on, but I can't seem to get rid of a certain grain. It is less with shadows turned off, but that is not really what I wanted. The sky is a starry background, btw. There's a reflective shader in the rock, so I thought that may cause some grain, but turning that off does not erase the grain.
It's only primary cloud, btw. With mpd 0.6 and AA6, and cloud quality on 4! I thought it shoud be rather smooth. Anything else you guys can think of?

Second is a card. I add some fractals to a surface shader, masked by simple shape and UV mask to keep the whites to the center, and with an opacity mask (default shader) to be sure, and also a non-reflective glass shader preceding the surface shader. Still I get this darkened rectangle. It does not seem the case in default render.
So, how can I get rid of that dark glass? I didn't change the RDM, btw, but that would only work on the details I thought, and the whites being smoke, it might as well be a little blurry. But not dark.

Some more information. I used the robust sampler, default settings, DOF is on, soft shadows by default too.
In the last row of tests, you can also see that PT does something dotty to the rock, which is not the case in the deferred render. I'm pretty sure it's the little bit (0.1) of (non-RT) reflection I put on the rock. But I do not like it. So I should probably turn that off if I want to render this in PT mode. Any thoughts?

Tangled-Universe

As far as I know TG still does not support grayscale-opacity, that would be my first guess when I see "soft square mask - object UV" as secondary input to the multiply scalar.
Complementary to this a IOR of 1.01 should render near transparent, but still might cause that tiniest darker tone. I wonder why you actually chose 1.01 there.

For the clouds I think you're a bit conservative with your render settings, especially when considering how dark the scene is. This is challenging for the robust adaptive sampler, especially with PT.
I'd set MPD to 0.5, choose AA8 default sampling, set cloud quality to 0.75 and cloud scatter quality to 300.
If that does not suffice then there are a couple of steps to try:
1) increase exposure, this helps the robust sampler detect noise
2) reduce PNT from 0.0375 to 0.02
3) increase max paths per sample to 64
4) increase cloud quality and/or cloud scatter quality

It's difficult to say which of these are most costly...perhaps you'd prefer option 4 over option 3.

Dune

Thanks, Martin.
The opacity mask was just added as an extra, to take off the (hard, I know) dark bits of the fractal at the sides (with the greys/whites centered), the rest of the smooth opacity inside the central greys should be taken care of by the glass shader. That was the idea.

I may well have been cautious with the cloud settings, not wanting huge render times. But the exposure idea is not what I had thought of, I can indeed always reduce it afterwards in PS. I did think about not using the robust sampler, as I found earlier that it's prone to grain.
I set MPD to 0.6, because I wanted the stars to be sharp enough, but I don't know if 0.6 cost a lot more, actually. Five might be good enough, I have to check.

It's a v2 cloud, so there's no cloud scatter quality afaik.

I set IOR to 1.01, because in earlier tests (long ago) I found that all rendered black if I went lower than 1. Maybe exact 1 should be better. Time for another test.

Tangled-Universe

Yes IOR<1 renders black. I recently had that too when attempting to render bubbles in water.
I'm still not sure if we are understanding each other correctly about grey scale values not working for opacity inputs?

My suggestion for MPD 0.5 is just because of rendering with PT, as it can save time, but it also depends on how much displacement you have going on. If it's not too fancy 0.6 may still perform well.
I don't have experience with starry background textures vs MPD settings. With defer all in combination with AA8 I'd tend to think that you render much more precise than without defer all and MPD..say...0.8.
The robust sampler has its strengths and weaknesses, I can understand your reasoning and choice. In many cases it's still the better sampler, but it needs some light to work with.

Yes cloud scatter quality does not apply to V2 clouds, so you can ditch that suggestion, but I think the other principles still apply.
Cloud ray marching quality for V2 is cheaper to render than for V3, at 1 or 2 should not hurt too much?
I think increasing exposure (to 3?) along with more paths per sample will already improve things quite a bit?

Dune

I think we are. I know opacity by default shader is blunt, so I just used it coarsely axe off the parts that were totally black, hoping that that part of the card was not rendered at all. But actually, I think it did render, so that was not necessary or doesn't work in PT. With the glass shader finer nuances of grey should be more or less transparent is how I thought.

For the most part MPD 0.5 would be fine, I know, I was just hoping 0.6 would yield sharper stars. I didn't compare though.

I found that a large quality (2-4) for V2 cloud doesn't cost too much more, but I think your suggestion to increase exposure is the one to try first.

Though I still have to get rid of the 'grain' due to rock reflections in PT.

Thanks for chiming in.

KyL

Quote from: Tangled-Universe on June 08, 2020, 12:15:41 PMMy suggestion for MPD 0.5 is just because of rendering with PT, as it can save time,

I wasn't aware of that!

Do you have any more details on this? I imagine that forcing a higher MPD would change the balance between primary and secondary rays, which would maybe explain the speedup?


@Dune Did you check your shader roughness value? Quite often those bright spots come from a high energy/small surface area. I also find it hard to work with the robust adaptive sampler in low light situation. Usually I go back to the old sampler, which provides cleaner results for the same amount of time.

Tangled-Universe

#6
I think the emphasis is on can KyL.
As you say it may affect the balance between primary and secondary rays.
Since the PT renderer is so much more expensive than the rasterized renderer it's best to try to shoot for as few micropolys as possible.
In many cases MPD 0.5 is a good starting point then.

Addendum:

If you are rendering a scene with clouds only and deferred shading (default with PT) and no terrain then you can lower MPD to 0.1 or 0.05 if you will.
The only object being affected by MPD in this situation is the background sphere, but that does not require highly subdivided surfaces.
From then on AA does the job for you.

This is why deferred rendering is great.
You need relatively low MPD to avoid facetted look of surfaces and AA does the sampling of the textures/shaders.
Without deferred rendering the 'shading rate' of the textures is coupled to the MPD setting.
For vegetation you usually like AA8 or even higher and with such AA values and MPD 0.5-0.6 the deferred renderer samples the shaders so accurately/densely that it's equal to non-deferred rendering with AA8 or over and MPD 1.5-2(!!).

Dune

Well, the AA 8, and PNT at 0.02 seemes to do the trick for the clouds. This part took 2 hours, so the whole thing will take maybe 10. Though I did swap the starry background for default again (may use an invisible plane with some fake stars or light sources popped on ;) , or something) and turned off the rocks reflections. I think the reflections in deferred render are softer (less conspicuous) anyway, because of the non-RT nature. With PT the real reflections might be calculated, strengthening the lights, thus resulting in a more grainy appearance (due to finer displacements). I will try a little part later on.
And also may try Kyl's suggestion of the older sampler, as I'd prefer keeping some shine on the rocks.

Tangled-Universe

Glad that that helped Ulco!
Yes reflections are softer with deferred and especially the PT. I'd reduce highlight spread/roughness until you have the same look. If you don't care too much about render times then the old sampler might give you more pleasing results for the reflections. I haven't really compared those directly for these kind of conditions to be honest.
I think this shows that sometimes developing a scene in non-PT with the goal to render it with PT can cause more work for the artist, but developing using PT is also much slower. Tough.


Dune

Thanks. Something I may use for the contest, top-right crop indeed. But I still have to decide if it's going to be moonlight or daylight, or something simpler, but more photorealistic.