Cloud Params

[treddie]

Howdie, all.

Was playing around with the cloud params in an effort to nail down how each param affects the whole.  I had plans on doing what a previous member had done; show examples of each change in a param with a sequence of images.  I do not recall who that was, but unfortunately, he did not cover the params I was working with.  Also, unfortunately, I found that as I built a sequence of images of these other params, I came away befuddled.  For instance, I ran two sequences...one with changes in only "Feature Scale", and one with changes in only "Lead-In Scale".  Someone had previously stated that "Feature Scale" determines cloud width, and "Lead-In Scale" is essentially the distance between clouds.  My tests put this seriously in doubt, as either param seems to favor the building of certain, random cloud areas while diminishing the size of other neighboring cloud masses.  In fact, if I were to animate both sequences, each would give it's own, random cloud creations over time, making the animation seem more like what you would see if you did a time-lapse photography of real clouds in action...certain areas building...others dissipating.  As a result, there is no easy way to build clouds.  For instance, I may have a set of cloud masses that I really like as far as positioning, but they may look "undeveloped" (not enough detail).  But when I try to add definition and contouring, changing the params gives me the detail I am looking for, but completely destroys the overall shape and positions of the clouds that I loved so much.  On the other hand, if I have clouds that are really great (good shape, position and detail) but are too small, there is no way to scale up each cloud about it's centroid.  I realize this is a perhaps impossible level of control over what is really a global, fractal phenomenon; the fractal equations determine the character of the clouds as a WHOLE, and to tell the equations to somehow "know" that THIS cloud over here is separate from this OTHER cloud over there, and deal with each as a separate phenomenon seems counter to the way fractals work.  Still, it would be nice to have more intuitive control over cloud formation.
<treddie

[nvseal]

I know what you mean. What I have noticed is that there is no one parameter that gets me what I want. You gave the example of having nice general clouds disribution but not having anough detail in the clouds themselves - I feel your pain  . There are certain techiques I have tried to get beyond this however. For example, If I have a really good overall cloud distribution but low detail, I might try lowering the smallest scale (therefore increasing the noise levels) while decreasing the color roughness (depending on what you are looking for, it could be the other way around). While this doesn't always work, it can be a good first step to getting wha you want. Another important aspect to take into account is how the cloud layer actually interprets the factal in 3D space. High density and low edge sharpness might mean less fine details in a cloud while lower density and higher edge sharpness might mean more details for clouds using the same fractal.
BTW it might nice to try to make a forum wide project to render examples of different values for the various parameter settings and have someone post them on their website. A visual guide to TG 2 so to speak. 

Edit: I think these are the test renders you mentioned (bottom of the page, "Atmosphere Examples") http://www.motionmagnetic.com/a_terragen2/a_terragen2.html.

[treddie]

Yes, those are the examples.  Thanks for posting that.

I will play with your suggestions.  Sounds like some good work-arounds.

I like your idea of a forum-wide version of the param thing.  Would certainly get it all done quite quickly.  The one thing I tried was to start out with a "reference" image, one that was always the same image (and settings) before the parameter I chose to work on was changed.  Except that kind-of didn't work...I quickly discovered that of the two params I tested, changes did not always occur at the same rate.  It was not as easy as incrementing the param by the same amount each time.  In order to get any noticeable change at all at some point in the sequences, I had to really exaggerate the size of the change, or go in the opposite direction and start reducing the size of the parameter, because working "forward" sometimes produced no changes at all REGARDLESS of how large I made the change.  So the idea of everybody using the same reference image to base their changes on doesn't really work, for, say all of those working on the cumulus cloud settings.  Which makes me think that what may result in quick results for a given parameter change for one person's clouds, may not result in the same response for someone else with different clouds.  Maybe the best we can all hope for are examples that APPROXIMATE what one can expect by changing any one parameter.

[nvseal]

Maybe the best we can all hope for are examples that APPROXIMATE what one can expect by changing any one parameter.

That's true. After all, the primary "look" of the clouds comes from the density fractal (so it seems). The same cloud density, for example, will have a different effect depending on the fractal type ( i.e. perlin or perlin billows). Giving examples of different density fractal settings would also be nice thing to try in such a project. Usually, the only comparison renders I've seen are those comparing cloud layer parameters rather than fractal parameters. I definitely think people would be willing to do it.

[Oshyan]

Feature Scale defines the overall median scale of cloud formations. The Lead-in scale more or less defines the "upper boundary" - the maximum possible scale for cloud formations. It does *not* control the distance between clouds. The closest thing to that would be something like the Coverage Adjust.

Each level or "octave" of noise (which are generated within the constraints set by Lead-in and Smallest Scale, and centered around the Feature Scale) is added to the others to form a more complex shape. The more octaves, the more "noise", so it's not just a matter of cranking up the octaves for maximum realism. You want enough octaves to be realistic while avoiding excessive noise and total randomness.

To control the scale of a noise function directly (cloud density shaders included) without affecting the number of octaves or shape, simply adjust the Noise Stretch XYZ on the Tweak Noise tab.

I think most of what you want to do is actually possible, and when you understand generally how things work it should also be reasonably intuitive. I don't think your parameter variation experiments are in vain and although there are some parameters (number of octaves, most notably) that would not animate smoothly, most will at least be instructive, even if they appear inconsistent.

- Oshyan

[treddie]

>Oshyan
Thank you for the info there.  It's great to have a good definition of the controls.  I will experiment.
It seems like "Lead-In Scale" is not a good term to use.  If I understand you correctly, it is essentially the opposite of "Smallest Scale", in other words the other side of the median.  Why not call it "Largest Scale"?

>nvseal
How about we start a new thread to that effect?

<treddie

[Oshyan]

I believe my explanation misses some subtleties in the specific implementation which make "Lead-in" a more appropriate term. Nonetheless a technically correct name is not as useful as a mostly correct and more well-known or easily understood one. The naming scheme is definitely something we want to look at.

- Oshyan

[treddie]

I hear you.  As a programmer myself, I have had to change the names of some of the things in the GUIs I build for various programs, because if I don't , I know perfectly well that if I don't use the program for about six months,  I'll come back and wonder what the heck I meant by this input or that.  Unless I use more user-friendly names that are a trade-off for more correct names.

I'm curious about the term "Lead-In".  What does it actually refer to?  Is it an aspect of a mathematical function?

>nvseal
My last post may have been misunderstood.  I meant we could start a thread about building a parameter example library.

<treddie

[nvseal]

>nvseal
My last post may have been misunderstood.  I meant we could start a thread about building a parameter example library.

I'm all for it.

[treddie]

>nvseal
We might want to hold off for awhile on that.  If Planetside is going to be changing things as they get closer to a final release, it sounds like there might be additions and deletions in the GUI controls, and names of things changing and all that.  There is nothing more frustrating than to go through all the trouble, just to go back in and do it all over again.

[Tangled-Universe]

>nvseal
We might want to hold off for awhile on that.  If Planetside is going to be changing things as they get closer to a final release, it sounds like there might be additions and deletions in the GUI controls, and names of things changing and all that.  There is nothing more frustrating than to go through all the trouble, just to go back in and do it all over again.

Good suggestion, I agree!

[rcallicotte]

Hopefully, not much will change, but I agree.  This sounds like a very good idea.

[treddie]

This would be such a cool reference.
<treddie