After experimenting with my locally run AI in early 2022, I found a model that was trained on the entire wiki-art collection. From there I tried to let this AI imagine moving renaissance paintings.

These newly imagined works feature ambiguous landscapes that are created by the machine flowing from shapes that include mountains, seas, bodies, and cities, without any distinction between the subjects.

The neural networks are trained on a collection of thousands of historical paintings, most of them dating back to the Renaissance. The networks do not possess any knowledge about the world on which the previous historical creators based their work. The generated works therefore show the essence of the database it was fed: it assumes the previous artistic translation as an objective truth and therefore not a subjective expression.

The historical paintings are strongly influenced by the physical world and the maker's experiences in this world. The historical makers reflected the physical world through reasoning, semantics and symbolism, strongly influenced by the technology of the time. The neural networks do not possess the capability of semantics or reasoning; they see the body as a landscape, the sky as a surface and mountain ranges as patterns.

No sound