CLIPFontDraw: Stylizing Fonts With CLIP

Abstract

In this study, we propose a method called “CLIPFontDraw” which enables a user to generate font images stylized according to text input without reference style image. By combining a large-scale multimodal model(CLIP) and a differentiable renderer, our proposed method generates a style matching a short input text and renders font images of individual characters in that style. Furthermore, our approach can also render simple logo patterns with the generated styles. The results of an experimental evaluation of our proposed method show that its output was more readable and expressed the input text more clearly than existing baselines.

Keywords

• Font style transfer
• font synthesis
• CLIP
• differentiable renderer