More Aladdin's lamp sort of testing on an older set up, by which I mean trying to use new v3 clouds to see how they look in place of the old. Also trying for a little virga effect.
Noble One by Walli
wagon by stectoons
xfrog
I like this a lot...........you have a little noise in the clouds IMO but that could be intentional. :D I've been doing a lot of the cloud conversions also, some have been successful, some not so much. :-\
Quite convincing downpour on the way there! Love it all, esp. the texture on wagon.
Echo others - well done!
I like this a lot...very American West....
Very nice, a bit ominous.
Somehow very creepy mood, but great picture anyway!
While trying to reverse aging of a family photo album, I did a little editing. You know what I mean, underexposed photos and such.
This render looks great in many tones, temperatures and exposures. :) This is where I noticed the foggy layer....great effect.
Thanks for all comments. Tried to render again to reduce the noise and as I had accidentally left out a wanted low cloud layer final density connection. The rain was supposed to appear to fall from a layer with a little more definition substance and density.
This is a composite of new with previous. The new render took 19 hours (8 hours in GI pass) but still noisy. Hoped to get rid of cloud grain/noise but it didn't happen and I lost more in the upper clouds than I gained, hence the composite. After all the time/work decided to post the combo anyway.
Defer Atmos was on. Used 16 atmos samples. Cloud layers (there are five of them counting some ground fog) all at quality 0.5 with acceleration cache set on Optimal instead of none. Render prepass GI detail 1 quality 1. GI in clouds set to still/very high. AA 8 Detail 0.8. Render Sampling 1/64 first - noise threshold 0.035. 19 hours ? Where did I go wrong ?
Matt said in a previous post that you probably couldn't tell the difference between still/medium and still/very high after I had a 17 hour render. That could be the problem. ;)
Thanks Yossam, I did some testing.
I set all the voxel buffers to 0.0 and the GI on still/low and a very small render took 4:04 minutes.
With Gi set on still/very high the render took 4:25 minutes which is roughly a 9 percent increase for such a simplistic case.
All tests with cloud GI still/low and five cloud layers defined.
With each layer voxel buffer set to 0 the render took 4:04 minutes.
With each layer voxel buffer set to 0.001 of a million (only 1000 voxels ) the render took 9:16 minutes.
With each layer voxel buffer set to 0.01 million (only 10000 voxels) the render took 9:43 minutes.
With each layer voxel buffer set to 0.1 million (only 100000 voxels) the render took 9:10 minutes.
With each layer voxel buffer set to 1 million the render took 12:59 minutes.
With each layer voxel buffer set to 10 million the pre-render stage alone now took over 15 minutes and the render took 38:35 minutes.
Test images showing the changing definition in the clouds.
More :
Only one cloud layer enabled - 50 million voxels - 2:48 minutes
Only one cloud layer enabled - 150 million voxels - 4:35 minutes
My idea here was to try 50 million voxels, but use that amount on only one layer instead of using five layers of cloud at 10 million each.
My conclusion is that multiple layers are very taxing on render time not simply voxel total.
Note the level of fine cloud detail in the 150 million voxel render.
I was just going to post a link where the cloud layer numbers were talked about.
But just saw that it was your post already :) Less cloud layers looks always faster.
Dear lord, don't use 0 voxels! Hah. New clouds need voxels to function. ;) A test with voxels set to 0 isn't really a valid test, basically (I'm not sure but there might even be an internal minimum since the shading doesn't work without voxels as far as I know).
Multiple cloud layers is definitely taxing, especially with the new cloud shading model. But I think there is also room for optimization in your settings to improve render times and quality without necessarily losing your 5 layers. Although I will say I can only really see 2-3 obvious layers in your original shot, so it may be better simply to work on getting the same effect but with fewer layers. Enable/disable each layer one at a time and see what is really contributing to the effect, and whether you can get a similar result with fewer layers. For example if you have 2 layers that together create a sort of "overcast" look due to their high overall coverage, instead of using 2 medium-coverage layers for this, you could use 1 more dense/higher coverage layer and get a similar result, probably with lower render time.
Anyway for some general advice on render time, first I would suggest not bothering with the Very High GI setting in clouds, especially for a scene like this that is frankly pretty "muddy" in the lighting already. Generally I would advise *using the defaults* and then optimizing/adjusting from there, so I wonder for example why you decided to use Very High GI for the clouds. Did you feel the detail was not enough with default setting of Still/Medium? It's also a bit odd to use Very High cloud GI with a main GI setting of only 1/1, while the default is already 2/2. I would not suggest going below the default, especially not with higher cloud GI settings.
To resolve the noise, it helps to know what the main sources of noise can be. Primarily they are Atmosphere and Cloud Layer shading, which are controlled by Atmo samples and Cloud Quality respectively, but are also affected by your Antialiasing settings. Note that GI does *not* cause noise, so if noise is your concern, increasing GI settings is not going to help at all. Instead, enable Cropping and start experimenting with your Atmosphere Samples and Cloud Quality. Generally I'd recommend a cloud quality of 1 or 0.75, I would not recommend less than that, especially with highly adaptive AA.
In your specific scene I suspect it may be dark and shadowed enough overall that the *atmosphere* may actually need more samples. I would suggest testing this carefully, find an area of high noise, especially an area in shadow, then increase Atmosphere samples to something like 32 and see if it is less noisy. If it's not, then atmosphere is probably not the source of your noise! So then instead take atmo samples back to 16 and increase Cloud Quality to 1. If that helps, the issue is in your clouds. It's easier to troubleshoot this with fewer cloud layers of course, and that's one more reason to start by testing with atmosphere since it's only 1 setting to adjust in a single node, rather than across 5 nodes.
I hope that helps you figure out the source of the noise and reduce render times.
- Oshyan
Some really good advice there, some stuff I was aware of and a lot I wasn't really conscious of...thanks Oshyan
Many Thanks for all the information Oshyan,
Keep in mind the intent of my initial render is as a test and way of learning what it takes to convert older cloud layers that worked well in T3 to similar looking layers using the new clouds of T4.
Obviously, the intent of using 0 voxels was a baseline test needed to establish what would otherwise be an unknown. Frankly I'm fairly surprised to see any cloud effects at zero, but there is certainly something produced other than blue sky :).
Having no access to the design or code being executed, I'm in the position of examining what amounts to a black box so bear with what seem like foolish tests to you. In first tests with unknown parameters I will always try zero or negative values in order to know what happens at the extremes.
The five layers from top down are :
1. high pinkish wispy layer
2. a yellowish wispy layer
3. darker gray cumulus layer
4. rain virga layer stretched in Y and also rotated to slant it
5. ground fog layer
I used cloud GI set at still/very high because it is the best available setting, I don't have pre-knowledge of what it will do (again these parameters are like a black box to me) , and my personal inclination is always to try out high settings.
I used GI prepass values of only 1/1, not because of noise but in an attempt to lower the horrendous prepass times I was getting which took 8 hours to run before even one final render bucket would even start to fill.
Another observation I may add is that the render has an old oil painting look to it. Those deep rich earthy tones.
Understood fleetwood, it's good to know the context of your experiments. A couple of responses:
I understand your inclination is to use "the best available setting"/high settings, but you also appear concerned with long render times. I would suggest *not* using "high settings" as a general policy, at least not until you have determined they might be necessary. If you do continue with the "use high settings" policy, you will inevitably experience not just long render times, but - perhaps more importantly - longer than *necessary* render times. Often times a high setting is useful or necessary for particular circumstances, but will literally produce no visual difference in rendered output in other circumstances, so you end up rendering for a long time for no discernable benefit. This is the cost of defaulting to high settings and not focusing on optimization and understanding how the quality and detail settings interplay. Of course I understand that testing can lead you to such an understanding, but I'd suggest starting with tests of lower quality first, at the least.
As for the GI prepass and your long prepass time, it's important to know that those settings do not affect cloud GI. If I'm not mistaken setting it to 1/1 probably won't notably reduce render time which is likely up at 8hrs due to the "very high" setting for *cloud GI*. Reducing cloud GI would probably give you the best improvement in GI prepass time.
- Oshyan
Thanks Oshyan for the ideas.
Really my main concern is almost always final image quality not actually render time. I accept long render times if I understand why I'm getting it and the end results are there. At this point with the new parameters I haven't tested enough t be at the "understanding why" stage.
I have the render (still noisy) down to about 5 hours now but have more tests in mind.