This Black Friday the new realme 7 5G at a demolition price: 229 euros!

Write something and watch this neural network automatically convert it to realistic calligraphy



Calligrapher is a small website where we can test the simple magic of a neural network that is capable of imitating human calligraphy in a more than convincing way. This site is an experiment in which we can enter text and see how an AI "writes it by hand".



The difference between something like this and other types of synthetic calligraphy generators, is that instead of creating a font that looks like human calligraphy in which each letter will always be the same, this site uses a neural network that has learned to copy human hand movements when writing lines of text.



The result is more natural and variable, more similar to how a real person would write





Screely 1605442463811

In this same image we see how each "Genbeta" looks slightly different



As instead of using a "freehand" style typeface to write the text, it is an AI that is "writing freehand", the text has variations like those that our own letter has when writing, with defects and changes in the strokes.



The web lets you play with nine different styles, change the thickness of the stroke, increase or decrease the readability of the text and the speed at which it is written. Too you can download your text as a vector graphic in SVG format.







What are GANs and how do they work, those neural networks capable of creating the faces of people that don't exist





Advances in artificial intelligence thanks to deep learning and neural networks are changing the way algorithms work so that they are able to get closer and closer to human perceptual power.



And when it comes to replicating human things, there is no shortage of experiments. Automatically generated faces, deepfake videos, conversations or stories, drawings that leave something to be desired, music with voices included, memes and even the news. The generation of freehand writing is not new, and this experiment is based on a paper by Alex Graves dating from 2013.



Via | sjv






Comments