Do you love drawing but hate colouring? Or perhaps you are a digital artist who is having trouble making the transition from paper to screen.
Then the PaintsChainer project may be just the thing for you.
Developed by a Japanese mechanical engineer with the Twitter handle @tai2an, PaintsChainer is a web service that adds colour to your line drawings at the click of a mouse.
PaintsChainer is built upon Chainer, which is an open-source framework designed for the research and development of deep-learning algorithms.
According to tech company O’Reilly, Chainer approaches computational graphs “on-the-fly” rather than in advance. This makes it a flexible and adaptable framework that can help users realise neural networks as easily and efficiently as possible.
PaintsChainer works both automatically or semi-automatically – you can either give it a suggestion for your preferred colour palette or just let it run wild on its own.
Artists – particularly in Japan – are already having plenty of fun with the software and have been sharing their creations with the world using the hashtag #paintschainer.
— elaine (@ms_manatee) February 2, 2017
— ivetto? (@_ivetto) February 4, 2017
— Kennit (@KingRhanTegoth) February 1, 2017
I thought I’d give it a shot, so naturally I found a sketch of Donald Trump and put PaintsChainer to the test.
I figured the software wouldn’t know the, er, colour scheme of the new US President, so put an orange dot for the skin and a yellow dot for the hair.
Neural networks work by simulating the human brain and using lots of interconnected computers, which act as “cells” of a larger whole. These computers can then learn things and recognise patterns almost in the same way we do.
There are many possible applications for this kind of thing. For example, these networks may prove useful for running complex systems like factories or air traffic control.
But, at least for now, we are mostly using them to mess with photos.
Remember in 2015, when Google’s Deep Dream AI was (terrifyingly) turning everything into dogs? That was done with neural networks.
And earlier this month, Google published a paper showing such networks could be used to restore pixelated pictures back to something more realistic. That might be super practical for finding criminals on CCTV footage, but could also raise privacy issues.
On a lighter note, a Russian team of developers has just released an app that uses neural networks to realistically edit your selfies, so you might want to have a play with that one.
You can try PaintsChainer free here, and share your attempts on Twitter.