Announcing F13

Today, F13 is coming out of stealth.

We’ve raised $5 million to build AI models for precise graphics generation.

AI-generated visuals have a reputation for being approximate. They can look impressive at first glance, but the closer you look, the more things fall apart: numbers are wrong, geometry doesn’t make sense, objects are drawn out of proportion, and so on. While often impressive and beautiful, they lack a lot and you can't accurately change them.

Images are one of the fastest ways to communicate information. A good visual can explain in seconds what might otherwise take paragraphs of text; as Karpathy puts it, they are a highway to the human brain. We believe the best medium to build those informative, useful visuals are vector graphics - the very same graphics that make up for your browsers user interface, logo of the brand on your clothes, design of the app you used today, or any presentation you've recently seen.

What makes vectors particularly interesting is their structure. A graphic is made up of individual elements with positions, dimensions and relationships that can be inspected, changed and reasoned over. That makes vectors a natural medium for models that need to understand not only what something should look like, but how it should be constructed, with an arbitrary level of precision.