Feedback Volumetric Clouds

The HDRP Tests project within the github repo does have a couple of scenes that use the volumetric clouds and those work for me. I have not been successful at getting it working in my own project/scenes yet despite switching on many options, but I didnt previously post about my failure to get it working because I need to spend more time trying first.

https://github.com/Unity-Technologi...Tests/Assets/GraphicTests/Scenes/5x_SkyAndFog

 

It's a little cumbersome to make them finally appear, as you have to enable them in 3 places.
You have to enable them in your hdrp asset, then in the default frame setting in the lighting sub-section and then you finally have to add them to your volume overrides and enable them there too.
https://github.com/Unity-Technologi...lumetric-Clouds.md#enabling-volumetric-clouds

Also increase the far clipping plane distance on your game/scene camera to atleast a couple of kilometers in order to properly see them.

If you still can't see them after all those steps, then it's likely a hardware/driver specific issue.

 

Yeah I reckon thats the one I was missing when I failed previously, because I forgot to change the 'Default Frame Settings For' dropdown to 'camera' when searching for all the places that had the Volumetric Clouds tickbox. When I first got this working a short time ago its because I basically achieved the same thing by different means - I ticked Custom Frame Settings on the Main Camera and then ticked Volumetric Clouds there.

 

Just download and run HDRP Tests. No clouds at all. Everything else is working.

P.S. Switch on Vulkan and clouds appear. Switch back on dx11 - no clouds. I'm on geForce 1650.

 

@pierred_unity still interested if there is a chance that planet position / radius parameters will be implemented for clouds system so it will be consistent with the PBR sky?

 

So, if I will set Earth curvature parameter to 1, but will change planet position parameter in the PBR sky, clk

So, I set planet center position in the PBR sky to Vector3(-6378100,0,0)

This is what I get



Why I set planet center position to this value? Because in the space simulators you can fly to the side of the planet. And as I can see currently clouds support only flat world case.

@pierred_unity so we should not expect, that this case will be implemented?

 

Also I set earth curvature to 1 and just zoomed out the camera and this is what I get:



As you can see atmosphere doesn't match with the clouds.

@SebLagarde is this and not respecting PBR sky planet center position expected behavior?

 

No you have to set it to 0 to match the planet's curvature, anything above that will make the cloud sphere smaller than the planet.

 

sound strange, have you tried to take a look render features in camera and HDRP Wizzard?

 

I don't know if it's a technical limitation, an artistic choice, or sensible performance settings in the examples I've seen, but the clouds do look a bit featureless and low-res to me compared to other solutions out there:




(source)

I think a lot of people will be wanting that amazing fluffy 3D cauliflower cloud look. But then again, I also have no idea how the cloud system from the gifs above performs so maybe it's not a fair comparison

 

Well, this is what I expect what the cloud system and pbr sky should combine into:
https://twitter.com/FewesW/status/1338497274482216964?s=20
https://assetstore.unity.com/packages/tools/particles-effects/overcloud-137742

 

But right now unity really glitches out when zooming out far enough for us to see some curvature of the planet

 

I gotta say, 2ms on base PS4 does sound excellent to me. I think having something like this is better for most cases than having "MS Flight Simulator"-quality clouds but having your game run like a slideshow

 

@pierred_unity in advanced mode, the Cumulus, Alto Stratus, and Cumulonimbus settings seems cannot be used together, only one of them can be applied. Is this intended?

 

I have some troubles authoring my clouds using the manual mode.
The red channel in the cloud map is supposed to define the coverage of the clouds, but it just changes the density of those (e.g. a value close to 0 makes the clouds appear like thin fog).

I would rather think that the density should be defined in the clouds LUT map, as this is the place where all cloud-type specific parameters are in.
This would allow to set the density for each cloud type in the LUT directly.

Currently I'm stuck to the cloud map's red channel in order to add density variation, but this has a major drawback since it applies the same density to all clouds at a given position (horizontal), so vertical density variation is impossible to achieve.

As you can see in the picture below, I've set up some types of cumulus and stratus clouds which have their own specific density, but due to the fact that I can't control the density depending on height, I had to remove other cloud-types above or below. This leaves ugly gaps in the cloud layers;


I hope there will be a solution to this in the near future.

 

Ok, I fixed it myself in VolumetricClouds.compute by changing it's code to this;
(starting from line 404, changes are marked with //MOD)

Code (CSharp):

1.     // Adjust the shape and erosion factor based on the LUT and the coverage
2. //MOD start
3.     //shapeFactor = lerp(0, shapeFactor * densityErosionAO.y, cloudCoverageData.coverage.x);
4.     //erosionFactor = lerp(0, erosionFactor * densityErosionAO.y, cloudCoverageData.coverage.x);
5.     float densityFactor = densityErosionAO.y * densityErosionAO.y;
6.     shapeFactor *= densityFactor;
7.     erosionFactor *= densityFactor;
8. //MOD end
9.  
10.     // Combine with the low frequency noise, we want less shaping for large clouds
11.     lowFrequencyNoise = lerp(1.0, lowFrequencyNoise, shapeFactor);
12.     float base_cloud = 1.0 - densityErosionAO.x * cloudCoverageData.coverage.x * (1.0 - shapeFactor);
13. //MOD start
14.     //base_cloud = saturate(remap(lowFrequencyNoise, base_cloud, 1.0, 0.0, 1.0)) * cloudCoverageData.coverage.y;
15.     base_cloud = saturate(remap(lowFrequencyNoise, base_cloud, 1.0, 0.0, 1.0)) * densityFactor;
16. //MOD end
17.     // Weight the ambient occlusion's contribution
18.     properties.ambientOcclusion = densityErosionAO.z;
19.  
20.     // Change the sigma based on the rain cloud data
21.     properties.sigmaT = lerp(0.04, 0.12, cloudCoverageData.rainClouds);
22.  
23.     // The ambient occlusion value that is baked is less relevant if there is shaping or erosion, small hack to compensate that
24.     float ambientOcclusionBlend = saturate(1.0 - max(erosionFactor, shapeFactor) * 0.5);
25.     properties.ambientOcclusion = lerp(1.0, properties.ambientOcclusion, ambientOcclusionBlend);
26.  
27.     // Apply the erosion for nifer details
28.     if (!cheapVersion)
29.     {
30.         float3 fineNoiseSamplingCoordinates = AnimateFineNoisePosition(positionWS) / NOISE_TEXTURE_NORMALIZATION_FACTOR * EROSION_SCALE_FACTOR * _ErosionScale;
31.         float highFrequencyNoise = SAMPLE_TEXTURE3D_LOD(_Worley32RGB, s_linear_repeat_sampler, fineNoiseSamplingCoordinates, CLOUD_DETAIL_MIP_OFFSET + erosionMipOffset).x;
32.  
33.         // Compute the weight of the low frequency noise
34. //MOD start
35.         //highFrequencyNoise = lerp(0.0, highFrequencyNoise, erosionFactor * 0.75f * cloudCoverageData.coverage.x);
36.         highFrequencyNoise = lerp(0.0, highFrequencyNoise, erosionFactor * 0.75f * densityErosionAO.y);
37. //MOD end
38.         base_cloud = remap(base_cloud, highFrequencyNoise, 1.0, 0.0, 1.0);
39.     }

The red channel in the cloud map now only controls the coverage, just like the red channel in the LUT map does.
The green channel in the LUT map now also controls the clouds density (this may was the intended behavior anyway but someone skipped it), so density variation is now possible at any horizontal or vertical position.

 

also can we get separate tilling and offset settings for each cloud type? having one parameter to affect all of them kinda limited

 

Really enjoying this! Couple callouts:

1. With volumetric cloud shadows enabled, there's a hard edge showing on the ground plane (image attached).
2. Additive VFX graph particles are not affected by the clouds (maybe more) -- It'd be good to know if this is a technical limitation.

+1 to point light support, +1 to "fly through" and "fly above" improvements. But this is already looking great!

 

For those of us working on Technical Animations or interactive products that do not require VR framerates, we would love to see much higher quality clouds like overcloud or simul truesky. Blobby looking clouds are far too gamey for archviz or realism work.

 

The cloud in unigine 2.14 looks good, hope unity could catch up them soon.

 

They aren't really affected by transparency either afaik (You can see this in the HDRP template where the butterflies are unaffected by specular reflection on the glass of glass box)

 

Btw could you share the coverage maps/settings for this scene, I'm not really a technical artist so, haven't been able to achieve anything like this

 

Hey I need to jump to my real job but I can give you a quick look at my settings (attached). The screenshot I posted earlier is pretty vague, but the clouds produced really came out amazing (moreso than I expected!). If I find time I'll throw a few texture maps together and post them here.

This is using the Advanced setting. I first tried to go the Custom route but I am going to hold off for the Unite presentation to completely understand the textures.

In my case, I used two cloud noise maps (coincidentally, that I was using for light cookies to simulate the lighting effects of clouds previously), and the cumulonimbus map is a third generic noise map. The two cloud maps are pretty generic photoshop noise generations with various gaussian level and brightness levels.

The cumulonimbus map and slider produced these incredible vertical columns. I was aiming for something that looked incredible to move through, rather than look up at (or down on), so being able to generate these was great.

This was set for a night-time scene, in a sprawling setting (large scale). I used a lower shadow opacity (0.33 and 0.33 respectively) for daytime. There are some issues w/ artifacts and edges showing up, so I mixed normal volumetric shadowing to get exactly the look I wanted. But fortunately it all holds up between drastically different lighting scenarios, which is what I'm always nervous about when getting this detail-focused on settings like this.

 

Following up @Ruchir here are some cloud textures I've been using in the Advanced setting. I don't know the relationship between texture size and the cloud map size in the settings, so it may be that these aren't optimized. Disable SRGB on import for these.

The "Clouds 256 Cells Soft" is the cumulonimbus map. The blown out whites are what is bringing out vertical pillars between layers as far as I can tell. I would probably blur this out a little in the future.

The "Clouds 256 Noise" Soft and Contrast are for the other maps. Contrast is what I choose for the rain map b/c it leaves more variation in the rain darkening.

I was looking for something wispy that worked well when broken up with craggy mountain-like terrain, when flying and looking down. I leverage the shape and erosion heavily and am not looking to have fully opaque clouds here. So, these may not work well for a lot of other projects, but the worst case is you can try them out and if you don't like it, at least you have a couple more tiling noise textures that may be useful somewhere else.

 

Thanks a lot

 

Anyone knows how these clouds would promote realtime GI(Enlighten)?? If I want to create a day and night cycle with weather changes will enlighten realtime GI be able adapt those changes?

 

If enlighten supports dynamic light cookies for directional lights, I don't see a problem with it

 

Also give us a way to create and assign custom volume noise, i take it the default cloud noise was generated using this tools? https://github.com/keijiro/NoiseTools , integrate and improve it to hdrp

 

I have a question though,
Shouldn't cloud rendering be independent of camera far clip, it should be a setting in the post-processing volume instead, because camera stacking is expensive in HDRP and to get satisfying result with volumetric clouds in any game they need to be rendered to quite a long distance, which will really kill performance on most hardware specifications if we were to just increase far clip of camera to get that

 

Can anyone tell me what kind of noise function is used in UE4?

The cloud noise looks much more natural here.

Also, recently I found this asset, even this does have better cloud noise:

Are there any updates in the works that improve upon this for the current volumetric cloud?
(Also, the cloud light scattering in UE4 looks Amazing )

 

Damn Expanse looks lit. . . .

 

@auzaiffe thanks for share. Really good things happening...

 

@auzaiffe

Great to see the improvements.

Do you plan to support reflection probes in the future? At this moment the documentation says that "Volumetric clouds do not appear in Reflection Probes".

BTW. Do they work for visual environment lighting input at the moment? I mean are they included in the ambient lighting?

Thanks.

 

The new settings and being able to flythrough without temporal accumulation artifacts seem pretty exciting

Also, one thing that could probably be useful in many cases is a halo effect for directional light, if I remember correctly Fog Volume 3 does it pretty well.

 

I don't know if I should post this here, but in the video is it that they are able to modify cloud height like a terrain, and it also act like conventional volumetric fog(I mean interaction with geometry, shadows and all)?

https://www.instagram.com/p/CPb79JxnYnU/?utm_medium=copy_link

 

Currently, I found two Bugs/Limitations in the clouds system:

• There is a strange directional light present when I set the sun intensity to 0 or there isn't one present in the scene. When I increase the directional Light intensity from 0 to say 0.01 the clouds darken and behave as they should instead of having an imaginary White directional light point downward.
• 
  
• 
• They currently don't take into account the Sky Emission or shy exposure value so, they look really odd. Black clouds in front of bright sky.
• 

 

Well, the default cloud layer doesn't react to lighting changes at all for most part so, can't really use it either:


As you can see with having 0 directional light intensity and very low exposure value for HDRI sky, the cloud layer is overexposing the whole scene and the stars in the background are just not visible

Even the clouds don't react to sky exposure values or color like they are always white no matter the skybox, They really standout during sunset and sunrise settings

 

Also the clouds are visible no matter my exposure settings for camera:




This doesn't happen when I have a directional light with intensity > 0

 

But it doesn't work with the exposure value for HDRI sky as well, which should work given I don't always want a sunlight in my scene, for example an gradient sky for a stylized look, where only the sky exposure value is used as ambient light.

The cloud should look consistent in this case and not pitch black imo.

 

Also, could you take a look at this?
See the 4th picture in that, it's not possible with a current implementation either (I think), the rain volume and the natural rainbow.