The curator's reviewA recurrent neural network with genuinely lovely handwriting. Type a sentence, watch it written out in real time, then slide the legibility and style dials and watch the same hand fall apart or flourish.
Calligrapher.ai wraps a 2013-era handwriting model - Alex Graves' recurrent neural network, one of the founding papers of sequence generation - in the simplest possible interface: type a sentence, watch a pen write it out in cursive, in real time, stroke by stroke. Sean Vasquez built the site; the model predicts pen positions the way language models predict words.
The magic is in the sliders. Legibility controls how disciplined the hand is; style switches between nine learned hands; stroke width and speed do what they say. Slide legibility high and you get a careful schoolroom script. Slide it down slowly and you can watch handwriting dissolve by degrees into pure gesture - loops that mean nothing, connected beautifully. Somewhere on that dial is the exact point where writing stops being symbols and becomes only motion, and you can hover there.
It matters historically: this network is a direct ancestor of everything in this collection. The idea that a model could learn a sequence so well it could continue one - pen strokes in 2013, words in 2019, worlds by 2024 - starts in large part here. That an artifact this foundational runs live in your browser tab, instantly, is a small miracle of preservation.
Type something you mean. There is an odd intimacy in watching a machine hand-write your own sentence back to you, hesitating at the same letters a person might. A tiny model, a decade old, still one of the most elegant things on the web.
