Neural Painting

Zahra Khosravi

I focused on using machine learning techniques in creating art. To be more specific, I used style transfer technique to apply the style of my art to other images. I created an interactive webpage to access the model that I trained on my style in the browser. I was inspired by creative projects that used machine learning as a medium and emphasized a sense of human touch.


I used style transfer to apply the style of my art to other images. This was a two step process. First, I trained a model based on my drawing and then I used my style to recompose images in the style of my artwork.

The style image that is used for the training process has to have certain characteristics. It needs to have very clear and repetitive patterns. The model is trained during a complex process that includes using a GPU on HPC services. I made the model accessible in the browser using Ml5.js library. I use the trained model on different input images. The model applies my style to the input images and outputs (generates) art. I repeated the whole process a number of times while I was changing different parameters in the code or training data set to improve the results. The decisions that I make for creating and picking the style image and input image are based on my personal experience as an illustrator. So far, I worked with famous artworks.The list of current input images is available on the website. I am interested in applying this model to works by my favorite artists. My target audience is an artist who is interested in using machine learning techniques to create art and wants to experiment with patterns and styles.