Case in Point: Machine Learning is a Practical Tool for Designers

I set out to check if Machine Learning can become a standard tool in a designer’s tool-set. Can it be used not just for eccentric art but also for commercial use-cases such as illustrations for websites, landing pages, application flows, and blogs?

In the first test that I ran, I showed that a machine can automatically color illustrations. It stayed inside the line and recognized skin, clothes, and hair and used appropriate colors for each one. The second step was to see if a machine can be trained to take a simple stick figure drawing and turn it into a more intricate illustration.

That second step was much harder. But after many trials, I got it to work. The machine turned lines into sleeves, pants, and necks; Elipses into heads; Rectangles into skirts. Results can be improved, but this is enough to prove a point, that a simple stick figure drawing can be turned, in the click of a button, to a colorful illustration.

This opens endless design possibilities. A machine trained on a few different data sets can instantaneously turn one simple drawing into many illustrations, each with its own unique style inspired by a design trend or an illustrator. Then multiple figures can be automatically compiled into complete scenes. Eventually, designers will use their creativity to inspire a machine that will produce dozens of design concepts.

Are you a data scientist, developer, or designer and want to team up and build an image-generating product? Reach-out, check my Medium profile for contact info.

References to illustrators whose drawings I used here: Katherina Limpitsouni, John D. Saunders, Pablo Stanley.

Reference to the algorithm implementation I used: Pix2Pix by Christopher Hesse.

Thanks to Zeev Kalyuzhner for ML advice and Aya M for help in creating the training data set.