Enunciating Pictorial Styles Via Generative Artificial Intelligence: Towards a Semiotics of Stereotypes

My article has a twofold objective: to understand the level of creativity we can attribute to generative artificial intelligences (GenAIs) such as Midjourney and DALL-E, and to describe the notion of enunciative style as it is transformed by these GenAIs.
It explores how GenAIs bring forth stereotypes about recognized artistic styles. My tests focus specifically on the styles of well-known painters contained in the databases. Indeed, when a prompt regarding an artist or an artistic movement is generated, the machine refers to all the images in the database associated with the words present in the prompt. The machine performs operations of abstraction and selection and thus generates images. By repeating the same prompt multiple times, different images are consistently produced. I quantify this variation of generated images in order to identify what I call “regional style,” meaning the patterns belonging to a specific region of the database activated by the given prompt. This regional style is distinguished from the “global style” unique to each model. Indeed, Midjourney, DALL-E, and Stable Diffusion, each developed on different datasets and systems, impart a specific style to all generated images. These experiments allow me to explore how generative models reproduce traditional visual arts by distorting and typifying them. In this context, by starting from familiar forms in artistic tradition, it is possible to estimate the level of stereotyping of the forms produced by GAI and also the level of creativity of the various models.