How to create Art using AI

3 min readFeb 29, 2020

A future filled with unseen proportions creativity with Man and Machine work together to reach new heights.

This has been an amazing journey of discovery and creativity.

Exactly one year ago I was on the phone with my friend and he was pitching me one of his ideas that triggered in me a side that I for some reason buried it deep. I still don’t remember why I buried it, nevertheless, my artistic side was alive again but it had been almost 10 years without actually making a physical sketch and on top of that I become consumed by Artificial Intelligence because I truly believe it can help solve humanity’s biggest problems.

Now, I have been learning everything I can about AI for 3 years and I have read so many papers such that if I got 10 dollars for each page I read I have close to a quarter of a million dollars. I’m not bragging by any means, most papers I didn’t even understand(you are not alone there) but I believed that if learn how to understand them I would hit jackpot. And indeed I did, I read a paper on style transfer that forever changed my way of thinking and actually gave plausibility to my gut instinct that gave me the intuition/feeling not to try and become Da Vinci myself, rather take a non-conventional approach.

Instead I decided to build myself a Da Vinci algorithm which in turn gave Birth to Beltrix Arts, although the core mission of my company has nothing to do with technology, nevertheless, technology is helping us shape our mission and bring it to life.

Enough of the back story, let’s get to the cool part the how-to.

Style Transfer

It is an interesting technique that separates the content and the style from an image, with the content being the structure of an image(your face in a selfie) and the style being the texture/brush strokes and colour of the artwork. Each of these representations can be extracted from certain parts of a Convolutional Neural Network.

The reference paper that I commonly see everyone go-to is the 2015 paper entitled: “A Neural Algorithm of Artistic Style” [1], a really great paper and amazing paper that judging by looks of it literally pushed the industry forward and inspired so many research directions. I have seen some Super-Resolution Techniques built on top of this paper such as: “Perceptual Losses for Real-Time Style Transfer and Super-Resolution” [2].

The main contribution of this paper is that the representations of content and style in a Convolutional Neural Network are separable. That is, we can manipulate both representations independently to produce new, perceptually meaningful images. [1]

I have gone through the paper and left my remarks. To inspire and help you read more papers I will leave the link to the marked paper and notebook so that you develop your skill or sharpen it.

Paper:

1508.06576.pdf

Edit description

drive.google.com

Notebook developed with Pytorch 1.4.0 and fastai v1:

Google Colaboratory

Edit description

colab.research.google.com

Thank you for reading, this has been the most fun discovery for me and the most fascinating notebook to develop right being Graph Neural Networks notebook.