Ekphrasis and Artificial Intelligence: Text-to-Image Generation in Theory and Practice

AI text-to-image generation tools, such as DALL-E, Midjourney, and Stable Diffusion, were introduced to the public in 2022 with great fanfare. This initial burst of excitement has subsequently been tempered by growing concerns regarding copyright, privacy, and the threat to creative industries they may pose. These multimodal artificial intelligence platforms require little to no specialized technical knowledge or equipment to use, which means that just about anyone can quickly generate a diverse array of images from text-based descriptions. Combining large language and diffusion models, these generators are highly flexible and, some have claimed, highly creative. The text-based descriptions used to create images on such platforms could be considered ekphrasis, but instead of describing an artwork or object that already exists, they invite users to describe images into existence. They actually create images by means of description. This session seeks contributions that both theorize the paradigm of image creation these new tools represent and describe their uses in artistic practice or academic study. What is at stake for art history and art practitioners in using multimodal text and image AI? Possible topics may include theory of the image, reflections on text and image study, history and theory of photography, issues of appropriation and authenticity, the use of such tools or the images created by them in cultural heritage or digital humanities research, applications in architecture and design, and reflection on art practice or discussion of artists’ work using text-to-image AI.