Script-component-based filter ideas

I am still hoping to do a 3D rotatable cube with image mapping some day...

Hulls for spaceships?
The existent filters are not very reallistic.
Cellular automats?
Chaos?
Real Voronoi patterns?
Uhhhhh.

i have no idea if this is possible with the new scripting, but how about something like sphinx built for making animations or even something that would automate that feature even more? that, being internal, would be really sweet!

I haven't seen any other raytracing examples but the one Vladimir had somewhere. Raytracing is out of my league. I might manage to get up to raycasting.

Variblur would be good. Could use height (alpha or selection) information for the radius, like Photoshop does.

IFS (iterated function system) fractals, very much YES.

?? What the heck do you use FFT/iFFT for?

Distance transforms: I'd be happy to be able to do just the basic Free Distort / Perspective Distort that is built in in PS.

FFT/iFFT: Once you're in frequency domain, it is very easy to do low/high pass filtering amongst other interesting things. But it is a rather expensive transform, which in the first place requires a doable bitmap based solution

About the IFS stuff.. is definitely high on my list too, but its a damn complicated situation trying to combine a projective design with a procedural

About the distance transform: its not a transformation in the normal sense, it will render the inner and/or outer distances of a selection or binary shape. I did work out a partial solution with the min and max components, but it is size dependant and very memory hungry. I guess it too requires a good bitmap solution

Next thing when I'm done with the 3D mandelbrot raytracer (lua is fighting hard back at me ) is probably cast shadows, i.e. a simple raytracer boiled down to the hard shadows part

Quote
CorvusCroax wrote: I think a Maze component would be cool.

You know, if you took ThreeDee's grid script and masked out areas based on a random table of 0s or 1s you can get a rudimentary maze that won't be solvable, but works.

I also write a python script for a game. It was in essence...

1. Create a random value via choice (0 or 1 or 2 or 3) create another random value set with only 0 and 1.

2. Create a MOD grid system (block out every other ups and downs and make each middle section editable)

3. Create the solution: given the condition that they must be continuous lines, if random value is 0 then turn left random value is 1 turn right (up and down so on). Save this on a 2D array. Do this until it hits the edge.

4. if the step already exists: Backtrack.

5. the rest: if the table is not touched fill the rest in. if 0 don't fill. if 1 fill.

Honestly the worst pseudocode I ever written but it works in my head. I wish I knew LUA better.

Quote
Deskar wrote: Hulls for spaceships? The existent filters are not very reallistic.

You should definitely check this out:
http://www.filterforge.com/forum/read...&MID=85737

Quote
Deskar wrote: Cellular automats?

Cellular automata are difficult to do with sample-based approach. The best approach would be to use a bitmap, but FF doesn't support bitmap-based scripting yet.

I'm working on a simple scripting component that I know I've been wanting several times, a switch for up to 16 inputs that is randomly controlled by the Global Variation, which input it will feed out. Will make a similar component for Curves as well. These will come handy when making filters where you want random controlled output or layers and to get rid of all "variation 1", "variation 2" sliders that are just random.

I would really like a script control that can be hooked to grey inputs, to generate true random sliders etc...

ThreeDee you asked for cast shadows so here they are...
Cast Shadows Filter

Quote
Jim Niehues wrote: ThreeDee you asked for cast shadows so here they are... Cast Shadows Filter

Oh yes, nice.

Hi Corvus,

Yes, it should. The Script random number generator works that way currently, it produces a different result every time unless tied to the global Variation setting. (I just hope it doesn't get fixed in the next upd ate )

And you gave me another idea that might be useful: A slider randomizer. That is, you could plug in a Script component that would randomize any slider within a given range (another input se ts the range). It would be similar to the built-in "Randomize" Variation functionality, but with full user control, not just the three preset levels of randomization.

Uhm, there IS a slight problem, which is that a Map Script can only generate an image output that plugs into a green input and the Curve Script a curve output that plugs into a blue input. Variation is a grey input, so you cannot plug a Script component there...

Quote
Skybase wrote: Watershed image segmentation?

Might run into memory issues...

Unless we get a bitmap component that you can run Scripts on, meaning if we could run a script ONCE for the whole image, not for every single render thread, it could be done. Currently you'd have to calculate the image as many times as there render squares. This makes some Scripts (at least) 64 times slower than they need to be.

This is really making some effects hard. It is painfully slow to do, for instance, an Internal Average Color script, for you have to calculate the whole image for every render square. No wonder it isn't in the Components. But blessed be the day a bitmap component appears there.

Hey here's another idea for a script:

Continuousness check / Flood Fill:

Basically just Photoshop's magic wand tool or the fill bucket, but in a FF node form. You give it an X and a Y and a tolerance. It returns black or white. (is continuous with point, or is not continuous)

I did some experiments in that direction, and yes you can easily make something simpler and faster than the bomber. Actually what motivated me is that noise octaves ("layers") in the noise components are blended and I wanted some sort of opaque style combining (its way faster than blending as you can stop sampling when you hit an opaque sample)

its not very complicated making such scripts, basically you need a table holding size, offset etc whatever you need to be unique per sample recursion. Then you loop n times where you fetch a sample and blend/mix/do your thing.

I'd be happy to help out if you need help constructing something like this.

interesting ideas shows above that could be done

Frequency-modulated random tessalation. That is, tessalation patterns which are more dense at darker areas of source image.

Beta skeletons