noise generators (forked from: Krita user community?)
mw_triad at users.sourceforge.net
Sat Feb 23 04:41:26 CET 2008
So, perlin noise is nice, but since I'm thinking about other noise
generators... I'm tinkering with a plasma generator (this one isn't
"strictly" procedural, although it might be possible to come close...
not that it should matter so much for krita the way it would in a 3d
render engine), and wondering if there would be interest in adding it to
krita at some point.
(There is a quirk in the generator that it can only make images that are
integer multiples of the seed spacing, so it potentially has to
over-generate... but the generated image, if not cropped, is tilable :-).)
All versions of this algorithm are based on an initial "seed grid", much
like perlin, that is then filled in using some algorithm. Currently the
KDE version only supports a "dumb" tesseral algorithm that fills one
cell at a time, independently (and is thusly a candidate for massive
multi-processing), which is why the no-jitter image is so obviously
block-like. A better approach I have uses a multi-pass fill that
recursively fills the midpoint of the largest-empty-space-containing
quads (this is necessarily multi-pass rather than tesseral because one
of the three stages for each pass depends on the nodes belonging to
neighbor cells). Currently I only have a VB implementation from many
years ago, but it still runs, so I can show pretty pictures. A benefit
of either algorithm is that they could cut passes short, i.e. a 50%
reduction in image dimensions (e.g. in a preview) results in a 75%
reduction in calculation.
The KDE version uses what is effectively an infinite-octave (or at
least, always down to whatever octave has a 1px frequency) multi-pass
where the opacity of each octave is O(n) = j * d**n, which lends itself
mostly to either traditional octave-passes, or octave-passes with
greater bias given to the higher octaves, effectively adding traditional
noise to the low octaves. The VB version (apparently) does something
quite different that more closely resembles an output filter (or the
aforementioned "stucco" algorithm) or a low-amplitude phase pass, and
looks much more "fractal".
My next goal after implementing the multi-pass fill algorithm is to
Something I would eventually like to get working is a fill algorithm for
non-rectangular regiions; at first quads, then triangles. At that point,
using a Fermat spiral for the seeding becomes an interesting
possibility. No idea if I'd ever get around to any of that, though.
Pretty images at
(The 'reuse seed' option in the VB version is apparently broken, so I
don't have similar images to show, also it just uses a dumb PRNG, not a
People say I'm going insane. I say, "what do you mean, 'going'?".
More information about the kimageshop