Glyce is a Chinese character representation toolkit that leverages visual glyph information to improve NLP task performance. It extends BERT to incorporate character-level visual features and provides pre-trained Glyce-BERT models trained on Chinese text. The approach combines visual representations of Chinese characters with traditional embedding techniques to boost performance across sequence labeling, sentence classification, semantic role labeling, and dependency parsing tasks.