I have to come back to this thread: https://planetside.co.uk/forums/index.php/topic,25484.0.html (https://planetside.co.uk/forums/index.php/topic,25484.0.html)
I can't get rid of the grain in water (volume density=0.5, color=0.3), so I'd like some advice. Perhaps I forgot something obvious again :-[
Here's what I tried so far, self-explanatory. All PT, and there's hardly any difference in grain with different settings. In the lowest, I forgot to mention that I set atmo samples to 64.
I'm afraid, you'll have to increase AA, if you're using the path tracer. I did some tests, and with AA of 12 the grain is quite OK, I'd say. But of course rendertimes are way longer. Just for fun I rendered it with the raytracer, and of course it renders almost instantly.
0.02 pixel noise threshold is still pretty high, actually. You are looking for <= 0.01
Even with clouds 0.02 - 0.025 which I used recently to demonstrate noise in clouds, still had noise.
PS we recently (probably years now lol) discussed how PT water is more grainy than the rest of a scene (including clouds) so it's probably accentuated here. I think it has to do with subsurface scattering introducing ray or sample loss, and thus more noise.
Perhaps PT water could use a AA multiplier. So you could multiply the current AA when it's rendering water. Or maybe a sample multiplier. Something that effects the rendering to improve result, but only on the water, so it doesn't drastically hit everything else being rendered.
Thanks for your time, guys. I'll try an even lower pixel noise threshold and experiment a bit more. But I think this is really a problem that needs solving, as it's a pity to need such high AA only for the water. Your idea, Jordan, would be a nice solution, if implementable.
Indeed RT is really fast, but not so good on the veggies. A little grain is acceptable, because it makes a render look photographic :P , but not too much.
I even tried a non displacable sphere for the water planet, but that didn't work (at all).
I did some experiments, but didn't find a suitable (less grainy) solution for PT water murk, while still maintaining a reasonably speedy render.
The sphere is obviously the most realistic option, but a (non-displacable) card certainly is faster. In these samples the water itself was colored by density 0.5 and color 0.3 (grey).
The fastest were water renders without integral murk, but with a preceding 50% (and 0.3 grey color) surface shader, again, faster on a card. But less sophisticated, and not even very much smoother.
Only with an AA of 12 did the grain get a bit less.
I would like to note you're using 25 samples for the PT. That really won't help things. Low samples will inherently create grain from density of samples per pixel, or whatever you'd like to call it or perceive it.
But yes, like I mentioned we really need a way to boost the quality of water alone, without effecting atmosphere, clouds, terrain, etc. This because of the inherent quality factors for PT controls. You can't get truly nice close-up water volume, or subsurface without extreme quality settings which hit everything.
The samples didn't do much if raised a bit, unless perhaps 81 or so. Didn't test that.
The grain is really bothering me, also in the glass of this building. Set at architecture, and AA=6. What settings would further be required to get grainless transparent windows? AA=12? Atmo raised (because of reflection)? Subsurface didn't do anything. Doublesided?
I didn't expect grain in the glass. Weird.
However, I'd say AA 6 is definitely not enough. Usually I render my (potentially) final images with AA 9.
Maybe you could create a GI cache file for the whole image, and render only the areas with water (or glass) with a higher AA. Most of the time, when you render with the path tracer, and there are no clouds in the scene, you don't even need a GI cache file.
Good ideas, Hannes, thanks. I usually use AA=6, which is good enough for 'busy' parts with veggies. This apparently needs more. I once learned that AA is best used with even numbers, but I don't know if it still stands and what for (I guess some sort of efficiency).
Ok. That glass looks terrible. What is going on there? I don't think I've seen it that bad before.
I will do some experimenting today. It's quite horrible indeed with default settings.
I am not very satisfied. It looks like even quite extreme settings don't eliminate all grain. All crops enlarged 200% for easier viewing. Titles give some specifics. I wonder what else to try.....
I just had an idea. Is the glass part of the model? I recommend to make it a separate part set to not cast shadows. I have the impression, that it's not the glass itself, that's grainy, but the stuff behind it, which is then of course somehow indirectly illuminated. And this indoor stuff is always problematic and noisy with the path tracer.
Of course with the glass not casting shadows it's still interior, but who knows? Maybe it helps...
Ah, that sounds like an interesting idea, let's try it! Thanks Hannes.
PT glass has shadows disabled for itself. How I did Mandos cockpit, and had direct lighting and shadows in the cockpot.
But it may be the darkness within the model itself. Light seems ambient on that side.
As far as I remember there's a difference, when you render it as a whole or with separate glass parts not casting shadows. I may be wrong, but I can't check it at the moment. Anyway, it might be worth a try...
I thought that was fixed with introduced or improved glass/volume. I can't see how PT subsurface, water, or [volume] glass would function right if it had [uniform] shadows and not depth. Volume density should be what's cutting out transmission within the volume.
I have a test rendering in default samples, AA 9 now with a little cube and window scene. One separated, other single object.
Actually there IS a difference. Look at the graininess behind the glass that casts shadows.
I do, but i also notice the glass on the right might as well not exist. No refraction or depth screening, or reflection. :O
Which brings us back to quality control. PT Glass WITH shadow is doing the physically correct simulation, which requires more samples/AA to bring out the realism.
I immediately notice in my tests that PT looks real, but grainy, while no shadow glass just looks fake and like ray tracer. All you see is slight ambient sky reflection, and that is literally it.In my test the way the sun is, it looks like there is no glass, basically, cause all the effects that should be there, are not working.Kinda sums up how PT volume can't work without shadows (or looks entirely different, whatever is happening).
Actually there is at least reflection (as you can see in the side window behind the door). The glass shader is set to double-sided/thin-walled, so I guess, since the glass is infinitely thin, there would not be much refraction in real life.
Yeah, if there is no light reflection on the surface of the glass, no-shadow PT glass doesn't even exist at all. It's only providing reflection, which can screen the glass in examples like yours where there is a solid ambient reflection.
You are literally just seeing floating reflections. The glass is actually completely transparent if it's not receiving reflection from sky or something. if you mix it ever so slightly with a transluscent default shader (not shadow. the heck) with like a glass blue, it at least looks like glass is there if no reflections are present.
In fact it also almost looks like that no-shadow PT glass actually kills PT lighting. Converts the internal scene to like a ray-traced scene. There is literally NO bounce inside of the no-shadow PT glass object.
PT with shadow object, you can clearly see the light is bouncing around within the object, and illuminating the interior.
If you aren't benefitting from the benefits from PT in things like this, it wouldn't even make sense to use PT. In this preview, lighting is set, so there are no reflections on the glass.
- Left is no-shadow glass separated.
- Right is a single object with glass material.
The right is doing the physically correct thing. The left is doing, pretty much nothing. Just a floating reflection shader, that also kills PT bounces. xD
Update: Also there is a weird line within no-shadow pt object that's not part of the mesh. Traced the line in second image. Oh! There is faint ambient reflection, and that is shadow from the right-hand side of the window sill.
Here is the gathered test scene if anyone wants to play with settings, etc.
Quote from: WAS on April 24, 2022, 12:11:19 PMThe right is doing the physically correct thing. The left is doing, pretty much nothing. Just a floating reflection shader, that also kills PT bounces. xD
Indeed. Quite something to report...
I'm on the fence. To me, it sort of makes sense that if it's not casting shadows, these features wouldn't work properly. Depending on how accurate the glass is, it should be refracting rays that pass through it. Becoming secondary, or whatever, but also having their paths probably modulated at different angles, hence the bounce inside to get that global illumination.
But maybe it should still work without shadows for direct rays, I'm not sure. But it seems to be acting as if it's not handling shadows plus rays.
Once more we have an interesting discussion :) Thanks for that. I'll test my (Dorian's actually) house with separated windows, as that has windows all around.
Since almost not a single model you find on the web can be used right as it is, I have made a habit of separating glass parts of an object each and every time and make the whole thing real world scale. I then work on the textures and save them as TGO. This saves a lot of time.
You got it absolutely right, I totally forgot about it. Separated glass and set to no shadows. Huge difference, and pretty good even with architecture and AA=6. Too bad the glass doesn't work on complete models and rules out shadows. I was under the impression that it did, actually.
Now just wait for the final model to complete this
Cool! I'm glad it worked. :)
Yeah, so am I. Thanks again!
The first thing I did, after I saw your new windows was to try your initial murky water file. Unfortunately your water sphere was already set to not cast shadows, so this solution doesn't work for water... :(
Probably because the water shader has integral murkyness, unlike the glass in the house.
It's working on completed models. It is not working on your separated object with no Shadows. Just reflection, and killing PT bounce within the house so I just see static direct lighting, which honestly you wouldn't see so clear from outside in the light. But meh. Might as well render SR windows that at least have some rake refraction screening, no? This is just broken. Most the windows don't even appear to exist due to lighting.And because there is still grain, you just have grainy screens in place of windows (like to keep flies out).
If you are going to separate objects, couldn't you just mask to the house and windows for a a higher quality PT? Object without volume will render fast, and then can get actual PT internals of the house with real bounce.
It's also weird that non lit trees are captured in reflections, but the building is. Only tree visible is the porch one, which reflects only shadowed object? I swear that's all you see on houses in relative shade, is all the sunny environment around, and lit trees.
In fact clear glass shouldn't throw shadows, so I don't see why it wouldn't work. But I didn't set double-sided, so I might try that as well (needed for internal bounce, I don't know).
This bit of grain can be taken out, but what I had before was really awful, even in very high settings!
From the settings I was doing, it takes really high, ridiculous settings. Hence the independent control being really needed.
And glass does throw a shadow, in fact, and that shadow is usually tinted with the shade of the glass, like that common green-blue. But I think the real issue is no shadow is somehow telling the renderer not to do any GI, which takes into consideration shadows.
But yeah glass does have a shadow as it does dampen rays passing through it to a degree, so when you say, take a sheet of glass over a sun ray or flashlight there will be a noticeable dim, and you can see the sheet of glass passing over the ray (shadow wise on the ground). It's even more intense with windows for houses, which use gasses between the windowpanes. They always seem to have the strongest effect outside automotive glass because of the plastic coating (safety glass).
I think, you're right, Jordan. Physically speaking glass casts shadows. But since CG is more or less an aproximation to reality, I could live with the not existing shadow of let's say a window glass. But the killed GI behind it like we could see in your screenshot is not good. However in Ulco's last image of the house it's not too obvious.