Computational creativity of neural network Midjourney in a polymodal space

Abstract

This article deals with the polymodal space in the field of computational creativity in neural networks. The object of research is a polymodal environment that integrates a series of heterogeneous codes to express a common idea, and the subject is the possibility of creating polymodal digital art using text and voice prompts in the generative network Midjourney. The aim of the study is to prove that computational creativity can be detected and described based on the results of iterations in the process of creating images, which in turn will allow us to talk about a complex polymodal system as a separate digital category of polymodality. We used the continuous sampling method when collecting linguistic units as they occur in the analysis process; contextual analysis for the systematic identification and description of the verbal and non-verbal contexts. It was necessary to conduct an experiment with the generative network Midjourney to identify patterns in the creation of a graphic space through text and voice data input, and then compare and contrast the results of iterations with the original image. The scientific novelty consists in the lack of research on the polymodal space in the context of neural networks and their generative ability. During the experiment, we obtained the following results: the term ‘polymodality’ in the context of the generative network Midjourney and its ‘digital art’ is due to the presence of three channels: verbal, visual and voice; tests have shown that the ability of the neural network to create images through prompt is at a high level, however, there are rough technical errors that do not allow users to fully approach the desired result when they generate an image; the summarization of the data allows us to talk about the presence of features of computational creativity in generative networks.