Game Development

3:41 PM

I have a small question about noises... is it possible to "stretch" the noise ? To increase the distance between high value noise patches ?

Im currently generating an forest, im totally happy with the forest patch size and the distance between them. However, in another map, i want those forest patches have more distance between each other... they should keep its size and form, but just have more distance to each other.

How can i do this ? Is there any setting that does this for a noise pattern ?

It sounds like you want to decrease the sampling frequency of your first few octaves.

So if you have a bit of code like accumulatedNoise += amplitude * GetNoise(x * frequency, y * frequency), use a lower frequency number for the first few iterations.

The higher the frequency, the smaller the feature size. When you're summing up a fractal noise (FBM) or turbulence, the first few octaves have the lowest frequency and determine the scale and spacing of the biggest, broadest features - where the mountains are, effectively. The later octaves with higher frequency control the size and spacing of the hills and bumps.

Thanks ! Im using an library for the noise generation... does this mean i need to come along my own noise generator ? Or is this math transformed on top of it ? Wouldnt it make sense to only use like 1 or 2 octaves if the first few have the lowest frequency ? ^^

4:03 PM

Using just 1 or 2 octaves will generally make your noise look very smooth and blobby / blurry, because it's missing the high-frequency detail. That might be appropriate for some applications, less so for others.

@genaray This depends on what your library offers. A reader who does not know what library you are using will not be able to answer this question, obviously.

Okay thanks, i guess i will start with a lower frequency and combine that with a only 2 octaves... i assume due to the smaller frequency the patches will appear less often. But therefore they are much bigger... so i think i should also increase the noise-threshhold for the trees spawning there... is this correct ? ^^

Again, you are asking me questions about how working with your library will behave without telling me what library you are using.

If you want a specific answer, ask a specific question. Show code.

Is this really library dependent ? I thought frequency and octaves behave the same way, or is this wrong ?

I'm making assumptions about what your library is even FOR. It might be using a completely different noise algorithm than the basic Perlin FBM I'd guess at. I've told you what I can based on that guess, and I'm not really interested in speculating further.

4:25 PM

As I understand it, octave is just a way to refer to a particular parameterization of Frequency, Wavelength & Amplitude. Most applications double the frequency & half the amplitude as the octaves get layered together, but that's not etched in stone.

@DMGregory The assumption was correct. I should have stated that...

@Pikalek Oh alright, thanks ^^

It depends highly on the context. In the generator I'm working on, frequency is only loosely set. You specify a minimum & the output for a layer (octave) won't be higher than that, but it will be lower in some places because of how the noise is generated behind the scenes.

Similarly, it could combine 2 layers with the same frequency & amplitude and the composite would effectively have about twice the frequency as most of the waves will crest between each other. That doesn't mean that's how all APIs behave, it's just a consequence of my implementation.

@DMGregory regarding the interior points thing - I agree that relative v interior wasn't well defined. Went ahead & posted anyway. Surprised that we don't have a canonical for that. Ditto for the recent questions about curve finding.

Yeah... should probably make one. They just tend to be so caught-up in the particulars of "how do I fit THIS data?"

Nov 25, 2021 at 17:29, by DMGregory

Thinking of canonical dupes, I wonder if it would be useful to make a canonical Q&A to link up all the "I need to make a formula that gives me these values" type questions - giving strategies for how to find formulae in general, teaching to fish rather than giving a fish.

4:40 PM

That's a good point - some of them could have overlapping answers, but not all answers are going to work for all situations..

For instance, not all will hit given data pairs exactly. Which isn't always needed, but sometimes it is.

Either I had forgotten or missed that chat. I'm going to go with missed & mixed in something about great minds thinking alike :)

If canonical isn't the best approach, is there a tag that fits (or could be crafted to)? If so, when we get something new, we could first suggest that they narrow their search to prior art to see if something already posted covers their situation.

Canonical might well be the best approach. I can imagine at least a decent canonical version, even if it's not perfect. I just haven't had the time to make a good one. ;)

Would it be an edit on an older question or would you add it as a new question?

4:56 PM

I think probably a new question, to separate it from the particulars of a specific data set to model.