More newlines than I expected. Looks good.
Thorn Fractal now has overload functions support for Custom formulas and it is used in alternating chaos series.
A small picture to demonstrate thorn fractal with overload function:
Now, I have added the legacy alternating chaos modes back. There is now 8 different alternating chaos mode. Furthermore, on the gui filter, you’d note that overload options do not show up, they will only show up where applicable i.e custom alternating formulas.
1 Like
@Reptorian, it seems that your ‘reptorian.gmic’ file is currently breaking the update script.
Not sure what you’ve done there, but could you please review your recent changes ?
Seems fixed now.
@David_Tschumperle : I do not remember putting gui in place of cli, but as you note, I was working so much on improving fractal filters and finishing up on Thorn Fractal. It was exhausting to improve Thorn Fractal to full flexibility.
That being said, I may attempt this - Cracked Floor Effect | Paint.Net Fans
Once welshblue at getpaint.net forum does a new concrete texture tutorial, I may finally update Construction Material Texture.
1 Like
Ok, started work on the new construction material. Starting with recreating the crack.
rep_turbulence : check "${1=32}>0 && ${2=6}>0" skip ${3=3},${4=0}
e[^-1] "Render fractal noise or rep_turbulence on image$?, with radius $1, octaves $2, damping per octave $3, mode $4."
repeat $! l[$>]
f. 0
+noise. 10,0
b. $1,0
-. {ia} abs.
^. $4
repeat $2-1
+noise.. 10,0 b. {$1/2^$>},0
replace_nan 0
-. {ia} abs.
^. $4
*.. $3 +[-2--1]
done
rm..
endl done
Does this version look better than builtin turbulence? I eliminated integer-based choice here as well as difference mode.
$ gmic 256,256,1,1 rep_turbulence 12,3.5,3,1.2
It is hard to tell because turbulence is, well, turbulence and random. After 10 comparisons of iterations, I would say that the builtin is tighter clustered than yours.
EDIT: Deleted earlier message as it’s no longer relevant.
New Filter! - Blur [Splinter]
Preview of new filter
Never mind about the question. I do wish ( ) can work with new lines rather than conjoined line.
Still very different. Maybe call it something else. dog example reminds me of the refraction and dispersion of gems.
What do you mean?
I wanted to create a 5x5 vector. So, I figured I could do
(1,1,1,1,1;1,1,1,1,1;1,1,1,1,1;1,1,1,1,1;1,1,1,1,1)
However, I think that would be so much better if it were possible to do
(1,1,1,1,1;
1,1,1,1,1;
1,1,1,1,1;
1,1,1,1,1;
1,1,1,1,1)
Also, making a new filter testing a convolution of 20 in the middle while -1 in the rest.
That will part of the new construction material texture.
(1,1,1,1,1;\
1,1,1,1,1;\
1,1,1,1,1;\
1,1,1,1,1;\
1,1,1,1,1)
That worked. Not on GUI code filter though.
Why is that? Do you have an example?
Yes.
Pinging @David_Tschumperle to let him know about this issue.
EDIT: I had to add two slashes to make it work?
Side note:
Any idea on how to optimize the Splinter Blur? I tried adding in conditionals to use convolve and convolve_fft depending on kernel size. Still too slow.
Sometimes you have to escape the escape. 
From David’s timing convolve (neumann) still beats everything. I don’t know his exact command and inputs. Am curious what your timings are for your system for your application.
I trimmed out the time to generate result by half.
https://github.com/dtschump/gmic-community/commit/26ff58b57ec59450139876440853057cab96849d
At this point it comes down to how convolve internally works. It still is quite slower than PDN version.
With the dog, I get 19 seconds with the new changes. Ideally, it’d be 5 s as in PDN.
In this case, I try to time the command without the convolve(fft) steps; i.e. generate an intermediary file to skip the processing. I do this all the time to separate the timing of the various parts. Without looking at your code, I bet there are ways to optimize your scripting a little more.
I’m on my search to a optimized version finally. I still don’t get how to get the min of value from here though.
Right side is optimized, and the left side is the non-optimized. I decided to loop within fill block to get a star, then use convolve_fft to get the image. It should take very little time to do the optimized version. And the value difference is because the left is normalized.
I believe that I can get the min or max using erode,dilate. Problem is that it has to be weighed, and I don’t know how to do that with a gradient star image. Any idea?
Here’s the code to work with
New Optimized Splinter Blur Convolution Map
#@cli rep_splinter_blur_convolve_map: _length,_duplicates,_thickness,_angle,_contrastr,_bisided={ 0=one-line | two-line }
#@cli : Create a convolve map for directional blur. This enables one to create a convolve map for one-direction motion blur.
#@cli : Default values: '_length=10%','_thickness=5%','_angle=0','_bisided=1'
rep_splinter_blur_convolve_map : skip ${1=10%},${2=3},${3=5%},${4=0},${5=0},${6=1}
repeat $! l[$>]
hypo={(sqrt(w^2+h^2)/2)}
if ${is_percent\ $1} length={round($1*$hypo)} else length={round($1)} fi
if ${is_percent\ $3} thickness={max(round($3*$hypo),1)} else thickness={max(round($3),1)} fi
rm
{$length},{$length},1,1
f "begin(
const sides=$6;
const thickness="$thickness";
const hw=(w-1)/2;
const hh=(h-1)/2;
rad2ang(v)=(v/180)*pi*-1;
const start_ang=rad2ang($4);
const dup_ang=rad2ang(360/$2);
rot_x(a,b,c)=a*cos(start_ang+dup_ang*c)-b*sin(start_ang+dup_ang*c);
rot_y(a,b,c)=a*sin(start_ang+dup_ang*c)+b*cos(start_ang+dup_ang*c);
cutval(v)=v<0?0:v;
maxcutval(v)=v>1?1:v;
);
endval=0;
xx=x/w-.5;
yy=y/h-.5;
lx=x-hw;
ly=y-hh;
radial_grad=1-sqrt(xx^2+yy^2)*2;
radial_grad=cutval(radial_grad);
for(n=0,n<$2,n++,
line=1-maxcutval(abs(rot_x(lx,ly,n))/thickness);
endval=max(sides?(line?radial_grad*line):(rot_y(lx,ly,n)<=0?(line?radial_grad*line)),endval);
);
endval;
"
/ {is}
if $5
avgstat={ia}
+f (i*2-$avgstat)
f.. lerp(i,i#1,min(1,min(abs($5),1)))
k[0]
fi
endl done
Does the normalization get in the way of optimization? It isn’t optimized if you aren’t getting the same or better result…