Shir Iluz*, Yael Vinker*, Amir Hertz, Daniel Berio, Daniel Cohen-Or, Ariel Shamir* Denotes equal contribution

A word-as-image is a semantic typography technique where a word illustrationpresents a visualization of the meaning of the word, while alsopreserving its readability. We present a method to create word-as-imageillustrations automatically. This task is highly challenging as it requiressemantic understanding of the word and a creative idea of where and how todepict these semantics in a visually pleasing and legible manner. We rely onthe remarkable ability of recent large pretrained language-vision models todistill textual concepts visually. We target simple, concise, black-and-whitedesigns that convey the semantics clearly.We deliberately do not change thecolor or texture of the letters and do not use embellishments. Our methodoptimizes the outline of each letter to convey the desired concept, guided bya pretrained Stable Diffusion model. We incorporate additional loss termsto ensure the legibility of the text and the preservation of the style of thefont. We show high quality and engaging results on numerous examplesand compare to alternative techniques.