Thanks to a free web app called kalligraf.ai, Anyone can simulate handwriting with a neural network running in a browser via JavaScript. After you type a sentence, the site renders it as handwriting in nine different styles, each of which can be customized with properties like speed, readability, and stroke width. It also allows downloading the resulting fake handwriting example in an SVG vector file.
The demo is particularly interesting because it doesn’t use a font. Fonts that look like handwriting already exist for over 80 yearsbut each letter appears as a duplicate no matter how many times you use it.
In the last decade, computer scientists have relaxed these restrictions discover new ways to simulate the dynamic diversity of human handwriting with neural networks.
Created by a machine learning researcher Sean VasquezThe Calligrapher.ai website uses research from a Paper 2013 by DeepMinds Alex Graves. Vasquez originally created the Calligrapher site years agobut has recently attracted more attention with a rediscovery on hacker news.
-
An example of handwriting synthesis on the Calligrapher.ai site.
calligrapher.ai
-
An example of handwriting synthesis on the Calligrapher.ai site with a different style.
calligrapher.ai
-
With readability turned off, this computer has terrible handwriting.
calligrapher.ai
-
With increased legibility, the letters become clearer.
calligrapher.ai
Calligrapher.ai “draws” each letter as if it were written by a human hand, guided by statistical weights. These weights are from a recurrent neural network (RNN), who was trained on the IAM online manuscript database, which contains handwriting samples from 221 people digitized from a whiteboard over time. As a result, the Calligrapher.ai handwriting synthesis model is heavily tuned for English language writing, and the folks at Hacker News have done so reported Difficulty reproducing diacritics common in other languages.
Because the algorithm that creates the handwriting is statistical in nature, its properties such as “readability” can be dynamically adjusted. Vasquez described in a how the readability slider works comment on Hacker News in 2020: “Outputs are sampled from a probability distribution, and increasing readability effectively focuses the probability density on more likely outcomes. So you’re right that it’s just a variation change. The general technique is called ‘fitting the temperature of the sampling distribution.'”
With neural networks now processing text, speech, images, video, and now handwriting, it seems like no area of human creative output is beyond the reach of generative AI.
2018 Vasquez provided underlying code which powers the web app demo on GitHub so it could be adapted to other applications. In the right context, it can be useful for graphic designers who want more flair than a static cursive font.
This article was previously published on Source link
