Fractomorph alien creater

[fractower]

The fractal algorithm spatially mixes multiple brots with different scales and orientations. It initially starts with 4 random configurations. The user picks the most interesting picture which is used as a base for 3 new pictures with mutations.   

[kram1032]

Really nice. Any links or something?

[fractower]

Nothing to share except pictures at this point. I would like to add it to one of the 3D packages at some point.

[fractower]

The biomorph algorithm proposed by Richard Dawkins assumes some rules or genes for drawing a picture.  The selection process is described in my first post. This will describe how the pictures are drawn and how the genes are defined.

We are all use to thinking of the Mandelbrot set being centred at the origin, but it is possible to translate, rotate and scale the set in the complex plane. For example



Where is the new origin and is a complex number that performs the scale and rotation. Some algebra steps have been skipped for readability.

The next step is to add another M set generator and come up with some way to weight the contribution of each.



A simple partition function which can be generalized to any number of M set generators is used




The complex values and are the genes which are mutated at each step. Each M set genrator adds 4 new genes.

This mixing produces a wide range of fractal effects that extend beyond the M set generators.

[Alef]

Is this like electrical sheep? Any neural networks aka memory of choices?

[fractower]

The program currently does not have any memory of previous choices. That is a good idea so I probably will add some history. I would also like to have genes for colours and eventually 3D.

 

[kram1032]

Oh, that's an awesome idea.
Electric Sheep with the main control being wether you like what you see or not.

I love such things but I'd have a request before you get started with that, if you do so:
Can you make it allow multiple metrics?
Like, besides "I like" and "I don't like", some categories that go into genres. - So you can generate, say, specifically scary pictures, or very assymetric ones.
The purpose of that would be to not head right into the next local maximum of quality but rather to drive the image past that into a more theme-based direction.
I'd suggest some emotional categories (like scary, funny, sad, ...) and/or some more structural ones (like symmetric/assymetric, colorful, fine, grainy....) alongside the typical "like it"

[fractower]

I am not sure how to quantify emotive phenotypes, but symmetries are relatively easy. Bi-lateral symmetry is particularly conducive to producing faces. In the symmetrical render below you can see all 4 of the M set generators used to produce the fractal.

I am having trouble getting the DE correct. It adds a bit of interest in some cases, but I would like the option of super smooth colouring as well.

[kram1032]

I don't mean you to somehow quantify them.
Rather, you make people judge for themselves. They click on what they feel is *emotion*est or something.
Additionally, if you want to make it really fancy, you could add some machine learning algorithm behind the scenes, which, based on your input, tries to learn what corresponds to what emotion.
It's just that usually with such approaches, they tend to converge too quickly. The final result tends to be fairly nice but still quite boring. By adding additional information, basically by doing multi-feature-selection, you allow for avoidance of too early convergence in some pretty but plain shape.

That last images is amazing!

DE sure isn't a trivial task on this one...