The article describes an experiment conducted during a class entitled Representation of a Material Object at the turn of December and January 2025/26 with the participation of students of electronic information processing. The experiment aimed to test how large language models cope with the interpretation of metaphors, which are semantic stylistic devices, and their transposition into images. The research material used in the experiment consisted of poetic and critical ekphrases of works of art. Ekphrases were chosen because, historically, the primary function of this genre was to evoke image-like effects in the mind of the recipient, known as imaginings or visualizations. The experiment was preceded by three stages. The first stage involved reviewing scientific works devoted to identifying the stylistic specificity of the ekphrasis genre. This enabled the identification of stylistic devices with polysemic properties that are characteristic of this text genre. The second stage involved entering a prompt into the image generator in the form of a selected ekphrasis containing ambiguous semantic means. Then, a group of students attempted to draw conclusions based on images obtained from the artificial intelligence model. The third stage involves attempts to change the prompt in order to obtain meanings similar to the interpretative conclusions that a human being might draw from a polysemous text. The experiment made it possible to identify and present the possibilities of interpretation and then the visualization of metaphors by artificial intelligence. A presentation at the conference will analyse the process and results of this experiment. This raises the question of whether AI-based image generators will transform the long-standing competition between words and images.