Deepfakes for scrawl: With handwriting synthesis, no pen is necessary

Abstraction: RNN web app synthesizes realistic varied handwriting without fonts

Key points:

• calligrapher.ai renders typed text as handwriting in 9 adjustable styles (speed, legibility, stroke width); outputs SVG vector files
• Unlike handwriting fonts, the model draws each letter dynamically with statistical variation — no duplicate letterforms
• Based on a 2013 paper by DeepMind's Alex Graves; implemented by ML researcher Sean Vasquez running RNN inference in-browser via JavaScript
• Trained on the IAM On-Line Handwriting Database (221 individuals, digitized whiteboard samples); heavily tuned to English, struggles with diacritical marks
• Described as "deepfakes for handwriting" — plausible synthetic handwriting indistinguishable from human scrawl
• Recurrent neural network learns statistical weights governing pen stroke sequences from real handwriting data

Connections: Deepmind · Calligrapher AI · Handwriting Synthesis · Recurrent Neural Networks · Generative Models

Source: https://arstechnica.com/information-technology/2023/01/computer-generated-handwriting-demo-offers-deepfakes-for-scrawl/