Text-to-Image Synthesis of the River Daugava: an artistic investigation

Abstract

The field of AI-generated art has recently burgeoned due to developments in a variety of image-generating neural network techniques. Typing text prompts to produce imagery, a process known as text-to-image, underwent significant performance improvements in late 2022, leading to an uptake in interest and popularity. This paper presents artistic research based around experimentation with image generation representing the River Daugava. Image synthesis outputs are compared when utilising a generative adversarial network (GAN) and the Stable Diffusion text-to-image model, and the importance of the prompt text to the Stable Diffusion results is explored. Information is also presented in terms of the technical aspects of image synthesis and text-to-image, and the ethical and artistic considerations that arise are outlined.