Convolutional Art is the name of my procedural art project. The goal is to create a way of generating paintings that are both interesting and unpredictably diverse. As you will see, this is not just a colour-swapping algorithm.
This post will showcase some of the newer pieces of art generated by my procedural art generator and describe a bit of the process.
Though a bit dark, this piece exemplifies the process of how the generator works. Random mathematical functions are applied to create different textures in the different parts of the picture. These textures then feed into a selection of pigments, which here are probably a dark petrol green, a light mint green, prussian blue, and black. Finally, the same methods also decide what way the colour should be applied, for instance, with long or short brush-strokes.
On top of this, procedural motifs are created using a drunken walk method. Here, the background is made dull, while only the path trodden out by the drunken walk is brought forth, as here, in full colours, the fire of a magical boar. All of that simply from randomly displacing a dot a thousand times and tracking its position.
It should be mentioned that the drunken walk is only a tiny part of the effect of the full picture, since those green/blue/teal colourations do not appear automatically. The drunken walk plays together with the rest of the random mathematical functions.
However, oftentimes, these kinds of methods do best when used more subtly, or like here, in combination with a mirroring effect that add a wonderful symmetric composition to whatever stark colours the rest of the convolutional network spits out.
If you look closely, you may find the random walk motif paved in golden bricks, but other times, more noncontrolled convolutional features come to the forefront like here. I hold that circles are incredibly powerful as long as not overdone - this picture seems to find a balance.
I also managed to create a compromise between the outlandish colours of the digital colouring method and the more subdued pigmentation method. This piece might not be a perfect example, a bit over-saturated, but all in all, it is a joy to see the interaction of different elements.
I chose to put this here since it seems a more directed version of the previous piece. I think, additionally, a circle was included. But apart from that, it is all guesswork. It is truly a thing on its own. Talking technically, everything - contrasts, composition - is much lower frequency than the previous pieces, a calm reminder that art can be beautiful without necessarily dazzling the viewer.
If I were to title these, this would be named Arterial. The brown spot is slightly offputting, honestly, perhaps because of the completely inhuman aspect, that its lower half is clean while the upper half is strangely mottled. That's another layer of the program, of course. But it rarely stands out like this.
I have a soft spot for black, and oh, the white stays pure, yet colourful. Truly a stroke of luck, this.
If you are interested in more, if you want a deeper insight into how it works, I have a
step by step tutorial up. If you want to try it for yourself, a pretty well-featured version has been ported to
OpenProcessing. Also, I can recommend
a single piece to represent the whole project.
