GlyphGen

The problem

AI can generate a wide range of visual options quickly, but each follow-up prompt can restart the direction instead of preserving what worked in the previous result.

That is a problem for type design, where small formal differences carry the personality of the face. Language alone can be too constraining for the next iteration.

Research

I looked at type design, OpenType terminology, expert font editors, and AI-assisted canvas tools. GlyphGen sits between one-shot generators and expert font software, where there’s a lower learning curve, but still control over output.

Three goals shaped the prototype:

1. Explore broadly, quickly.
2. Preserve design intent across iterations.
3. Let the designer choose how much control to take on.

Interaction model

I designed GlyphGen as a node-based workspace, where all ideation and iteration takes place on a canvas.

GlyphGen supports three levels of control. Designers can delegate the next step, guide the model through structured choices, or edit the form directly.

The descriptions and scores help the user calibrate their reasoning, and let the tool auto-iterate.

Reflection

The ideation process is not linear, so branching and version control are central to the tool. Keeping briefs, specifications, images, and edits on one canvas makes the exploration visible and portable for designers.

The next step is extending single-character exploration into a system for the rest of the typeface.