Artificial Intelligence, Goya and learning to draw. First steps

Abstract

We have taught (or trained) artificial intelligence to draw after a selection of drawings and engravings by Goya, using the open-source Stable Diffusion platform. Our goal has been to find out what we can learn, as visual arts teachers —a very different topic would be as visual artists—, from the drawings created by artificial intelligence. We have tried variations of learning using different sets of model images, either many and diverse, or few and very homogeneous, including long descriptive texts accompanying each image, or only images without associated written text, and, above all, insisting on successive learning sessions with the same model images. We present a selection of the most satisfactory results: 40 drawings grouped in ten series of four.

For professionals in the teaching of drawing, the first three sensations with artificial intelligence are: speed, wide dispersion of the artistic quality of the results, and remarkable stylistic congruence with the drawings used as models. Our inquiry has been a two-way journey: artificial intelligence has learned to draw after Goya's drawings, generating results of acceptable quality, and we have learned to incorporate artificial intelligence into the learning of drawing.