High fidelity font synthesis for CJK languages
Summary
zi2zi-JiT is a conditional font synthesis model designed for Chinese/Japanese fonts, using a Pixel Space Diffusion Transformer approach built on JiT. It combines a content encoder, a style encoder, and a multi-source conditioning mechanism to transfer font style from a reference glyph to a source character, trained on 400+ fonts with rigorous evaluation. The project provides end-to-end workflow including data generation, LoRA fine-tuning, generation, and licensing details.