On generative algorithms • Hacker News Board

Re: On generative algorithms

#41

Post by RandomInteger4 » Fri, Sep 21, 2018, 4:01 PM UTC

aw3c2 wrote: post #3

Warning, this series of fantastically written articles on generative art might make you feel very small and sad compared to those who are capable of not only creating but communicating beauty like this. In other words, I highly recommend you try some generative art yourself and see that at its core, it is not that hard and YOU can do it too!

They feel like a more relatable human if you follow them on Twitter where you can see some of the ideas they're testing out:

@inconvergent: https://twitter.com/inconvergent?s=09

It's easy to forget that art and creativity is an iterative / scientific process when you just see the finished products.

Re: On generative algorithms

#42

Post by gmiller123456 » Fri, Sep 21, 2018, 6:03 PM UTC

bduerst wrote: post #30

aw3c2 wrote: post #3

Warning, this series of fantastically written articles on generative art might make you feel very small and sad compared to those who are capable of not only creating but communicating beauty like this. In other words, I highly recommend you try some generative art yourself and see that at its core, it is not that hard and YOU can do it too!

Any idea on what they used to generate patterns like this? https://img.inconvergent.net/img/plot/43adc57.jpg

There seems to be two pieces to generating that particular image. The first appears to be just a sphere with lines drawn at random points on its surface. The second piece are the "motion" lines that tend to point away from the sphere's direction of motion.

The first piece seems simple enough. Pick and random point on a sphere, and draw a few lines at random orientations a very small distance from that spot with no correction for persepective.

The second piece, the motion lines, seem to be a bit more complex. But they appear to be done by choosing a random spot on a circle's circumference, choosing an angle at random which is heavily biased towards the tangent line at that point on the circumference. Choosing a magnitude (length) for that motion line. And drawing some dashed lines along that path. Then repeating some new lines along that same path but shifted a random amount perpendicular to the motion line.

[I doubt I explained that second part very well, so don't try to re-read it if you didn't get it.]

Re: On generative algorithms

#43

Post by kragen » Fri, Sep 21, 2018, 10:01 PM UTC

sarbaz wrote: post #19

The Hyphae image (and the trees) have a result that's very similar to Diffusion Limited Aggregation, but the method of generation is very different. IMO the Hyphae way is actually less cool than plain DLA. The algorithm had a lot of degrees of freedom that were selected randomly, while DLA has very few degrees of freedom and gives a cooler shape I guess the point is that sometimes a simpler algorithm can give more in…

DLA is super awesome, but if you're starting from a point, it takes a lot of computation. Or, rather, what we thought of as a lot of computation around 1990. Hoff's version of "Hyphae" is orders of magnitude less computationally demanding.

In a lot of cases, if you're willing to throw a massive amount of computation at a morphogenetic problem, you can get a higher coolness-to-algorithmic-complexity ratio.