Fontjoy

Font pairing is one of those decisions that looks simple until you are staring at 1,500 Google Fonts with no principled way to narrow the field. Fontjoy reduces that decision surface to a single click. The tool presents a three-slot pairing — typically a heading, subheading, and body candidate — generated by a neural net trained to balance visual similarity and contrast. Lock any slot you want to keep, regenerate the others, or override a slot manually to steer toward a specific typeface. The preview text is editable, which means you can pressure-test a pairing against real copy before committing.

The differentiating feature is the neural net’s framing of the pairing problem. The docs describe the goal as selecting fonts that share an overarching theme yet offer pleasing contrast — a balance that is hard to codify but that the model approximates through learned font vector relationships. The GitHub repo is publicly available under MIT license, so the underlying approach is inspectable and the project is forkable for teams with specific corpus needs.

Fontjoy fits cleanly at the discovery phase of a design project: it accelerates the shortlist from hundreds of candidates to a handful of credible combinations in minutes. It breaks at every subsequent phase. There is no weight or optical size control, no dark-mode preview, no palette integration, no export format, and no way to save or share a session. Designers who reach a candidate pairing copy the font names manually and continue the process in their design tool of choice.