This project examines how AI image-generation systems transform the relationship between language and image. In contemporary generative media, users increasingly create images through prompts rather than direct visual manipulation. This shift raises a central question: what happens when words become the operational mechanism through which images are produced? The project approaches the prompt as a contemporary, inverted form of ekphrasis: the classical rhetorical practice of vividly describing images in words. Whereas traditional ekphrasis uses language to evoke a mental image, AI prompting uses language to generate new visual outputs through algorithmic systems. The project therefore asks how meaning is preserved, transformed, or lost when text and image are repeatedly translated into one another.
The research focuses on Percy Shelley’s 1819 ekphrastic poem On the Medusa of Leonardo Da Vinci in the Florentine Gallery. This poem is especially suited to the project because it already involves a chain of mediations: Shelley describes a painting attributed to Leonardo da Vinci, which itself depicts the mythological figure of Medusa. The Medusa myth also resonates with contemporary anxieties about AI images, including deception, proliferation, and the possible displacement of human creativity. Methodologically, the project combines computational image generation with close reading and visual analysis. It will generate images from Shelley’s poem across several AI platforms, including DALL-E, Midjourney, Stable Diffusion, and Adobe Firefly, and use visual clustering to identify recurring motifs and compositional patterns. It will also reverse the process by converting the historical Medusa painting into text descriptions using image-to-text systems, then generating new images from those descriptions. Finally, it will trace iterative translation cycles—painting to text to image to text—at multiple stages in order to examine visual and semantic drift over time.
The project’s contribution is threefold: it rethinks ekphrasis as a critical framework for understanding AI image generation; it develops a computational-humanistic method for analysing generative media at scale; and it offers new insight into how algorithmic systems reshape visual meaning, aesthetic interpretation, and the cultural imagination of images.
picture: Head of Medusa (attributed to the circle of Otto Marseus van Schrieck, c. 1600–1650), via Wikimedia Commons. Source: Wikimedia Commons file page