Documentation
Font Garden is one open font collection, not a catalog of families by named designers. Every glyph in it is drawn on one grid system, graded by a person, and used as training data for the open-weight models that draft the next glyphs. This guide describes the files, the engineering rules, the build and test tools, the data pipeline and the way contributions enter the collection. It is a draft, started 2026-11-04, and will change as the system does.
The guide is split into short chapters. Read the chapters you need for the stage you are at. Each chapter carries one of these labels:
- startRead first if you are new to the project.
- must→A requirement. Every source in the collection follows it.
- learnBackground and reference material.
- teamProcess notes for people who grade and merge work.
- nerdTechnical detail on the data and the models.
- templA template you can copy.
The structure of this guide is adapted from the Google Fonts Guide, published under the Apache License 2.0.
Introduction: the basics
What the collection is, the position it takes on authorship and licensing, the font format it is built toward, and the tools.
- start Post-authorship and open weights
- start The collection
- start The NeuralType format
- start Setting up a working environment
Sources: the files of the collection
The source tree, the license files and the hosting practice that the build, the checks and the corpus tools depend on.
- templ Start a font project
- learn From a specimen image to drafts
- learn Hosting sources on GitHub
- must→ Source repository structure
- templ Licenses
Engineering rules: grid, structure and grades
The requirements every source in the collection meets. They are stricter than ordinary font production because every glyph is also training data.
- must→ The dyadic grid
- must→ Outline structure
- must→ Glyph grades
- must→ Proposals and grading
- team Corrections and drafting evaluation
Production: building and testing
How the OpenType fonts are compiled from the sources, checked and proofed.
- learn The fgtools command
- learn Build the fonts
- learn Quality assurance
- learn Proofs and media
Data and models
How graded sources become a corpus, how the models are trained and evaluated, and what is released.
- nerd The collection is the dataset
- nerd Models
Contributing
How drawings and corrections enter the collection.