YuzuMarker.FontDetection

Why we included this project

Recognizing the exact font in an image with Chinese, Japanese, or Korean text is a hard problem, and this project is one of the few open attempts at it. It trains a CNN to identify CJK typefaces and extract their visual style, which helps when you need to reproduce or match a font rather than just read the text. The repo also includes a pipeline for generating synthetic scene-text datasets, mixing real backgrounds with thousands of fonts and multilingual corpora, so you can retrain the model on your own data. A hosted demo on Hugging Face Spaces lets you try it without local setup. If you work with subtitles, scanlations, or archival documents, the pretrained model and data-generation tooling are directly useful.