On style and mere stylishness in AI-generated art

This article offers a critical reflection on the problem of pictorial style in artworks made with generative AI, in particular, so-called text-to-image models. As I understand it, there are three factors that complicate the status of style in artworks created by this means. The first problem is taxonomic, and stems from a tendency to equate artistic style with an artwork’s signature features. The second problem arises from the difficulty of translating visual qualities into words, or the ekphrastic translation problem; hence, it is one thing to describe in words ‘what’ a picture represents, but it is another thing to describe ‘how’ a picture represents. The third issue is more critical than technical. A picture generated from textual input will be ‘in a style’ and ‘show a style,’ but it will not have a style of its own. Because, I argue, a visual art style cannot be formed from words, a text-generated image will exhibit ‘mere stylishness,’ a combination of manner and signature.