Generating African Artistic Styles Using Textual Inversion

In recent years, there has been significant development in text-to-image models, which combine language and generative image models, to generate images based on text prompts. In this paper we analyse textual inversion using an existing model in the context of African images. We use the Stable Diffusion text-to-image model for two main tasks. First, we generate detailed images conditioned on text descriptions. Second, we train the model to perform textual inversion using African contextual images. When provided with a text prompt, the resulting model can generate images in the style of the images used during training. Specifically, we present an approach to guide the creation of a text-conditioned image representing or evoking a style or concept using only 3–5 images. We learn to represent this concept through new “words” in the embedding space of a frozen text-to-image model. These “words” can be composed into natural language sentences, guiding personalized creation in an intuitive way.