Feedback Volumetric Clouds

Thank you @ElevenGame that helped!

I was finally able to get it going with this code below. It's super simple, but now that I know how to grab the Volumetric Clouds I can develop a nice weather system.

Code (CSharp):

1. using UnityEngine;
2. using UnityEngine.Rendering;
3. using UnityEngine.Rendering.HighDefinition;
4.  
5. public class CloudControl : MonoBehaviour
6. {
7.     [SerializeField] VolumeProfile volumeProfile;
8.     VolumetricClouds vClouds;
9.  
10.     public void Sparse()
11.     {
12.         if (volumeProfile.TryGet<VolumetricClouds>(out vClouds))
13.         {
14.             vClouds.cloudPreset.value = VolumetricClouds.CloudPresets.Sparse;
15.             Debug.Log(vClouds.cloudPreset);
16.         }
17.    
18.     }
19.     public void Cloudy()
20.     {
21.         if (volumeProfile.TryGet<VolumetricClouds>(out vClouds))
22.         {
23.             vClouds.cloudPreset.value = VolumetricClouds.CloudPresets.Cloudy;
24.             Debug.Log(vClouds.cloudPreset);
25.         }
26.  
27.     }
28.     public void Overcast()
29.     {
30.         if (volumeProfile.TryGet<VolumetricClouds>(out vClouds))
31.         {
32.             vClouds.cloudPreset.value = VolumetricClouds.CloudPresets.Overcast;
33.             Debug.Log(vClouds.cloudPreset);
34.         }
35.  
36.     }
37.     public void Stormy()
38.     {
39.         if (volumeProfile.TryGet<VolumetricClouds>(out vClouds))
40.         {
41.             vClouds.cloudPreset.value = VolumetricClouds.CloudPresets.Stormy;
42.             Debug.Log(vClouds.cloudPreset);
43.         }
44.  
45.     }
46.  
47. }
48.  

 

(the following takes into consideration expanse in a whole with its own features/tools you have to use with it: Expanse Fog, Atmosphere, and volumetric clouds.
Then the Same with HDRP using it's own volumetric fog/clouds/atmosphere.)

Problem with expanse is it does much worse on lower-end cards, like GTX 1060-70. Where HDRP's volumetric clouds are usable with these cards, costing around 1-1.7ms, compared to expanse's over 3 - 6.5ms even with reprojection which makes it look quite blurry unfortunately but you have to use it on at least 2, most likely 3.

Expanse's atmosphere is not cached, so every frame you are spending 1.5~ms on it, changed or not,
where in HDRP you spend nothing.


I do agree that HDRP volumetric clouds could do with some visual improvements still.
Especially considering many people want to use Volumetric clouds for cinematic, animations, archviz, and in general projects that don't care about FPS, you offline render them so in the end they look smooth, even with 1 realtime FPS.


In expanse you can get crazy visuals if you don't care about FPS, in HDRP volumetric clouds, there's not much you can do to what's already there, even if you're willing to sacrifice a ton of performance.

Expanse fog is also quite powerful and versatile, and I feel can look better than unity's in some places. But in the same time unity's fog is more artist friendly and easier to use.

In 2022 there will be a full water solution which i very much look forward to, that hopefully plays well with volumetric clouds, it's going to be really nice.
Hopefully volumetric clouds get some love as well, polish here and there, if the author has time.

I feel like further work/improvements on unity's volumetric fog can benefit the volumetric clouds a lot.

In summary both are great, HDRP Volumetric clouds performance is really great, authoring is amazing, visuals aren't bad, but I feel like it could be better to be honest.

Expanse visuals are downright insane, authoring is pretty good, but performance can be really heavy on anything lower than an RTX 2070/super, especially on GTX 1060-1070 which many people use still. And if you want the clouds to be performant you have to make sacrifices in quality with reprojection frames and other settings for fog/atmosphere.

 

Hello All!

Is there a setting where you can define world scale? My project uses 1/10 in scale, a decision that was made years ago based on another cloud solution that now comes back to haunt me :-(

Kind regards,

 

Hey everyone, once again super flattered to see so many positive things being said about Expanse on this thread. Extremely blessed to have folks using it and enjoying it so much.

I've been hard at work on performance optimization, and am happy to report that on my 2080 Ti, I can get Expanse to render decent looking clouds somewhere between 0.3 and 0.5 ms. To PutridEx's point, I have no idea firsthand how this scales to lower end hardware... because I have none, haha. I can't think of anything that would make it absolutely tank if you are ok with lowering the noise texture quality to something your GPU's bandwidth can handle, though again, I don't have the resources to profile myself.

Here's a cloud layer rendering in 0.35ms.



Here's a different sort of looking option also rendering at 0.35ms.



Ultimately I'd bet that the HDRP clouds still edge it out---one of the reasons for this is that I lose 30us copying the motion vectors from inside a fullscreen fragment shader due to a compatibility problem in 2020.3 (rip). All excuse-making aside, my recommendation is still to go with the builtin clouds if performance and hardware scaling are top priority for your project. Pierre has also posted some great looking references in this thread, so I think it is possible to get them to look quite good.

I think if we worked together we could probably migrate most of Expanse's features to Unity's HDRP implementation with little to no performance impact, and I'd be fully on board with a deal in that department. That said, I'd guess that Pierre + team have other plans for what they'd like to do with this system.

Btw, anyone who wants to chat about clouds (Expanse, HDRP, or even... in the real world?), feel free to drop by the discord: https://discord.gg/F3VQ2vJy9p

 

Has anyone got an idea whether the clouds' Shape Offset can be used for floating origin cases?

What are the units used as offset based on as -1, 0 and 1 give the same result?

 

I'm looking forward to the noise addition
Looks amazing.

As it is, the clouds are still really good visually. They add some really nice visuals, and are especially helpful for open world games with time of day. And the performance is amazing.

I took this video of forest scene I'm working on, just adding the clouds with a few clicks gave it a really nice touch, using one of the presets
It's running in under 1.3ms with a GTX 1070

- for anyone using the clouds, it's important to increase the volumetric fog distance, it adds a lot to it.

 

IMHO,just need one small step to make people satisfied.

Unity Team has already done a good volumetric clouds system,but they forget one thing.
"People judge it only by final results."

if you just look at the volumetric clouds parts in those AAA games,they are same quality as the volumetric clouds in hdrp.sometime they are even worse.
but the texture based clouds in background helped a lot to hide the bad parts,they done a great help to add more realistic details to the sky.

something like the Cloud Layers in hdrp.

but as the Cloud Layers in hdrp dose not react to sun light and ambient light,you can not blend it with volumetric clouds in dynamic lighting scene which is the reason people using volumetric clouds.

the Cloud Layers is simply useless.

volumetric clouds just need a texture based cloud layer in backgound which will react to sun light and ambient light just like volumetric clouds.

look at all screenshots post in this long thread,they all are suffering for the detailless sky background.

just one small step Unity Team could make people stop crying for the clouds in AAA titles, do not let this last step to ruin it.

 

Your test / demo scene looks very nice. Is this available for download anywhere to have a closer look and testing performance etc.?

Keep up the good work!

 

I disagree, about you being interested in constructive feedback, but okay, I’m bowing out of this thread.

 

OK to make sure I'll just ban you from this thread, so that next time you can talk about the lightmap uvs in the lightmapping forum, and you can talk about clouds, well, too late for that. So make sure criticism never becomes an angry rant and we'll all be good.

This was in response to reports.

 

Hey @pierred_unity, do you have anymore insight/ workflow tips that you may be able to share on the author process for the custom cloud LUT and cloud MAP?

I'd be really interested in studying the other cloud methods more.
Simple > Custom is probably my all time favourite right now.

I saw your 'unity' screencaps and read the docs saying that it wasn't a trivial task.

I also saw that one of the channels appeared to have some 360 camera input data, so curious if there's anything a-kin to volumetric fluid Sim capture or even photogrammetry methods in there somewhere.

Thanks in advance

 

Hi,

I encounter problems with cloud shadows not casting shadows on vegetation rendered with Vegetation studio pro and vegetation engine. A single prefab works fine (black grass model in the picture below).


For testing purposes I added a second directional light with a cookie and here it worked (also single prefabs).


May it be, that the shadow cookie is inserted to late, after vsp rendering?

Thx,
Skandy

Update:
I guess it has to do with Lightprobes not working with instanced indirect rendering. Mabe it works with using LPPV:

Found this: The light probes structure lives entirely on the CPU side and for each object that uses light probes that structure is queried to provide an interpolated value. That means that Graphics.DrawMeshInstancedIndirect doesn't work well with it, as all the per instance data needed for interpolation queries lives on the GPU side (and not on the CPU side). The best approach is to make sure that the light probe data is on the GPU as well, and that you can do by using LPPVs. I see that you wrote that you tried that, however let me just provide the details of the needed steps, so that we're on the same page. The Graphics.DrawMeshInstancedIndirect docs specify that the last two parameters are Code (CSharp): Rendering.LightProbeUsage lightProbeUsage = LightProbeUsage.BlendProbes, LightProbeProxyVolume lightProbeProxyVolume = null Please use Rendering.LightProbeUsage.UseProxyVolume and specify which volume to use by providing a reference to it. See the "Global Illumination and GPU Instancing" section in GPU instancing docs. This LightProbeProxyVolume can live on any GameObject. However in your case the size and position of that LPPV can't be inferred automatically (as that automated logic needs a MeshRenderer). Thankfully you can use a custom bounding box that will encompass your instanced meshes. See the "Bounding Box Mode" section in Light Probe Proxy Volume component docs. The above will make sure that the data is interpolated on the CPU for all the vertices of the LPPV, passed to the GPU and in there it will be interpolated per pixel for each of your instanced meshes. Does VSP+VE support LPPV?

Now everything is fine.

 

I have one more suggestion @auzaiffe , could you add some more options to clamp and remap the shadow cookie used for directional lights?

I want these controls for clouds like these:

Where normally the cookies will simply be black (or whatever the max shadow opacity is), but I want to mimick overcast sunlight with some variations in it, so it would be really useful to have clamp and remap parameters

 

In short, yes

But I can imagine a curve could be expensive, so just the top-level controls for remapping would be really helpful

And, Happy New Year

 

Hey,

I encounter issues with the clouds:

Physically based Sky and Clouds

Sun a bit lower, physically based sky and clouds

Sun at same height but clouds off

Sun a bit lower, physically based sky and clouds


Is there a fix I'm missing?

Edit:
Found the issue with my light color temperature having a minimum at this time of day.



Best,
Andre

 

@auzaiffe ,are there any query APIs on the C# side available to integrate clouds into our gameplay?

Like finding density, visibility, etc.

 

Yes, some use cases could be:

• Finding if clouds have a high enough density so we could add rain effects. (Or something like Ace Combat 7)
• Calculate if the sun is visible at a spot for a farming simulator or any kind of simulation.
• Calculate visibility between two points.

Expanse has good documentation about it: Gameplay Query Interface (bguesman.github.io)

 

Hi,

I would be glad to have a cloud shadow intensity multiplier, or is this physically incorrect? I have reflection probes, lightprobes, lightmaps baked and all indirect multiplier are 1. Exposure is set on auto between 9 and 14. I don't want to set the lower exposure bound lower, as the shadowed areas without clouds would be to bright then.

Cloud shadow

No cloud shadow


Thanks,
Andre

Edit:
For anyone interested, i reduced the cloud shadows here.

 

I think that's already available as shadow opacity, no?

 

Hey,

okay, you seem to have a newer version. Good to know

 

not newer version.. click on the top right button visible in your screenshot & enable "Show additional Properties"

 

These settings are available by enabling "Show Additional Options" by right-clicking- the Volumetric cloud component.

or just click the three dots beside the "?" icon.

 

Hello
do clouds have a global texture we could grab for the shadows to use in shaders
I would like to use it as a mask in shaders for rain and other effects

 

https://github.com/Unity-Technologies/Graphics/pull/6828

Some improvements are coming to the Cloud Layer.

I'm wondering if Volumetric Clouds will support thinner cloud types like cirrus in the future or if it's optimal/better and 'by design' choice to use Cloud Layer with Volumetric Clouds together?

 

Is there any way to access the visibility/density of the clouds from a set location by casting a Ray or similar?

 

I would like to cast up from the terrain to get the cloud cover at different locations and create a Texture that can be used as a mask to drop rain/snow. density would be nice to know so you can deposit rain or snow depending on the cloud density. Hope that makes sense. I am doing this successfully with Expanse Volumetric clouds and Atmosphere but would like to extend to using unity volumetric clouds

 

That is right

 

There's an issue with a 2 directional light setup (sun + moon).
When the moon becomes the main light (cast's shadows), the lighting of the clouds still uses the sun's light color and intensity.

This can be easily fixed in HDRenderPipeline.VolumetricClouds.cs around line 395 as follows;

Code (CSharp):

1.                 if (!shadowPass)
2.                 {
3. //mod swap ... use the actual dominant light values
4.                     //cb._SunLightColor = m_GpuLightsBuilder.directionalLights[0].color * settings.sunLightDimmer.value;
5.                     cb._SunLightColor = currentSun.color * currentSun.intensity * settings.sunLightDimmer.value;
6.                     if (additionalLightData.useColorTemperature)
7.                         cb._SunLightColor *= Mathf.CorrelatedColorTemperatureToRGB(currentSun.colorTemperature);
8.                 }

Bug report;
https://fogbugz.unity3d.com/default.asp?1399000_te3908pvu2270oi9

 

Here you go;
https://github.com/Sebastian9112/2-Lights-Cloud-Test

I noticed a different behaviour when I recreated the setup in a new project.
The moonlight is not even affecting the clouds lighting until you push the light intensity to like 10000 lux and then it suddenly pops in.
There are two scenes in the project, a simple 2 light setup and another scene with a daytime controller which handles the blending/shadow-casting of the lights automatically.

 

That's fast!

 

To prevent people getting furious?

Sorry

 

Very punny

 

Hey there! Just out of curiosity, but are there any future plans for baking snapshops of the Volumetric Clouds into a texture that can later be authored as a Cloud Layer? I would love a high-performance alternative for projects that do not rely on dynamic clouds while at the same time offering an in-engine tool to author custom Cloud Layer textures.
Ontop of that, there might be an incentive to develop high-quality features, like the aforementioned multi-level noise, that may be too computationally expensive for most target platforms yet be a good candidate for baking high-quality clouds.