Scene text editing has gained widespread use in fields like poster design. In this study, we propose a new task for scene text editing that keeps the image background and text content unchanged while only modifying the style of the text. To accomplish this, we developed a web-based application that enables style transformation of text in scene images using prompts. As shown in Fig. 1, our application successfully transforms text in an image into any desired style using prompts. Experimental results demonstrate that the proposed application can naturally apply different styles to scene text without altering the image background and text content. Furthermore, our model does not require any additional training and is capable of real-time style transfer for scene text.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

SceneTextStyler: Editing Text with Style Transformation

  • Honghui Yuan,
  • Keiji Yanai

摘要

Scene text editing has gained widespread use in fields like poster design. In this study, we propose a new task for scene text editing that keeps the image background and text content unchanged while only modifying the style of the text. To accomplish this, we developed a web-based application that enables style transformation of text in scene images using prompts. As shown in Fig. 1, our application successfully transforms text in an image into any desired style using prompts. Experimental results demonstrate that the proposed application can naturally apply different styles to scene text without altering the image background and text content. Furthermore, our model does not require any additional training and is capable of real-time style transfer for scene text.