A little-known application of artificial intelligence (so far) is to analyze a pictorial language. When a neural network process images, it analyzes shapes, colors, directions, textures and compositions. It is a way of stripping a language of all symbolic references.
A network can be trained to do exactly the opposite, and find objects, people, or whatever, on a image, but that is out of the scope of this experiment.
To analyze images in purely pictorial terms, stripped of all symbolism, a StyleGan2 neural network has been trained on a set of about 300 images of my paintings. These paintings have been specially selected looking for a certain visual continuity, so I have focused on examples from 2017, 2018 and 2019.
After processing the images, we obtain a model, that is capable of reproducing infinite variants, all of them free of symbolic value.
Cracks
The cracks pattern comes from the not painted green lines around objects.
Emerald green and fluorescent pink
The use of this range of greens, and the occasional use of fluorescent pink has been common in my career since more or less 2014.
Leather and fabric
.
Simplified landscape
In my paintings, the characters usually appear outdoors, surrounded by vegetation. The landscape does not usually participate in the scene, and therefore it is usually limited to simply establishing a horizontal line that creates an illusion of space..
