How AI is Revolutionizing Chinese Font Design: Inside Alibaba’s Smart Font Lab

Background

AliMama, Alibaba’s marketing platform, launched the “Smart Font” initiative to provide free, high‑quality Chinese fonts for commercial use, reducing licensing costs and streamlining design workflows.

AI‑Generated Fonts

AliMama Number Black (Shuhei) – 6,767 Chinese characters, 52 Latin letters, 206 punctuation symbols; optimized for e‑commerce and advertising.

AliMama Number Smart (Shuzhi) – a brand‑standard typeface balancing a lightweight tech aesthetic with readability.

AliMama Oriental DaKai – derived from the Yan Zhenqing Da Bao Zi algorithm, preserving traditional calligraphic strokes while modernizing structure.

AliMama DaoLi – inspired by the Cun Bao Zi stele, featuring bold triangular cuts and “goose‑tail” terminations.

Technical Foundations

The font creation pipeline follows a human‑machine collaborative workflow: designers produce prototype glyphs, AI models generate the remaining characters, and experts refine the output. The core AI model builds on few‑shot font generation (style‑transfer) techniques such as DG‑Font , which separates content (glyph skeleton) from style using dual‑branch networks and AdaIN‑based fusion.

To overcome DG‑Font’s limitations, the team introduced:

Multi‑base‑font content synthesis that combines skeleton information from several base fonts.

A novel loss function that strengthens skeleton topology, improving transfer for highly stylized scripts.

Stroke analysis is performed on the OTF version of Number Black. Each glyph is decomposed into vector commands ( M, L, C) and classified (start/end points, outer/inner angles, closed shapes). This enables automated skeleton extraction and stroke‑decoration.

Open‑Source Font Toolkit

Alibaba released the font‑toolkit library ( https://github.com/alibaba/font-toolkit) which can load font files, query glyph paths, and automate tasks such as skeleton extraction, stroke modification, and font subsetting.

Font Subsetting and Variable Fonts

Because Chinese fonts are large, AliMama provides an online subsetting platform that preserves ligatures, vertical layout features, and variable‑font axes while reducing file size to a few kilobytes. The service ( https://www.iconfont.cn/fonts/detail?cnid=a9fXc2HD9n7s) generates CDN‑ready font files from a simple text input.

Variable‑font technology is also explored, allowing continuous adjustment of weight, width, slant, and other axes to create adaptable typography for different screen sizes and light‑dark modes.

Dynamic Typography and Future Directions

The project investigates dynamic fonts, including kinetic typography, physics‑based layout, and animated glyphs, aiming to enrich visual storytelling in advertising and emerging media formats.

Future work will focus on three pillars: variable‑font design, kinetic typesetting, and font animation.

Key Resources

GitHub repository: https://github.com/alibaba/font-toolkit Online subsetting service: https://www.iconfont.cn/fonts/detail?cnid=a9fXc2HD9n7s DG‑Font publication: Yangchen Xie et al., CVPR 2021.