next »

[gimymblert]

My first intuition to make it real time is to use a random distribution of point that is even in frequency (blue noise) but also value, and just threshold the value based on lightness.

[JobLeonard]

Yeah that's a well known fast approximation technique - remember those wang tile papers I shared before?

The thing about this one is that it finds an energy optimum using Voronoi cell relaxation. The process creates emergent structures in the way the dots align, especially if you combine it with a bit of edge detection in the input. It is really hard to approximate.

The devil is in the details here. You really need to let the simulation run for long enough for all of these structures to pop up - it can be a matter of single pixel alignment for each dot, but somehow the brain picks up on that. Better first guesses do help reduce the time to reach that stage a lot though.

Here are some screens. Most are details from high-resolution renders of Ernst Haeckel's "plate 53: Prosobranchia" (Kunstformen der Natur, 1904) that I linked below it.














https://blindedcyclops.neocities.org/haeckel%20stippled%20-%20variable%20size%20white%20dots.png
https://blindedcyclops.neocities.org/haeckel%20stippled%20-%20variable%20size%20black%20dots.png
https://blindedcyclops.neocities.org/haeckel%20stippled%20-%20variable%20size%20blend.png
https://blindedcyclops.neocities.org/haeckel%20stippled%20-%20variable%20size%20sobel%20biased%20white%20dots.png
https://blindedcyclops.neocities.org/haeckel%20stippled%20-%20variable%20size%20sobel%20biased%20black%20dots.png
https://blindedcyclops.neocities.org/haeckel%20stippled%20-%20variable%20size%20sobel%20biased%20blend.png