AI-powered image processing has brought incredible developments to the field image creation and manipulation. Deep learning techniques and generative adversarial networks (GANs) have been at the heart of models that create realistic images of fake people, turn sketches into realistic landscapes, and transform photos into the style of children’s book illustrations.
In a similar vein, AI Gahaku has been creating a buzz across the net for its ability to take pictures of people and make them into renaissance art. The system is simple: upload a photo of yourself with your face clearly visible, and the AI artist takes care of the rest. The result is an artistic rendering of your photo in a number of different styles. It’s a simple, quick, and fun way to explore a creative use of AI technology, and you can try it out here.
To learn more we got in touch with the maker of AI Gahaku, a full-stack developer based in Japan called Sato. In this interview, we talk about the inspiration and development of AI Gahaku, the challenges involved in making it, and how Sato learned to build an AI model from scratch.
Do you work in the field of AI technology, or are your AI applications a hobby?
I don’t work in the AI field at all, and usually I am just doing housework around the home. Programming is purely a hobby.
How did you get started with AI and programming?
I used to work at a bakery, and I started programming because I was interested in how different tasks were being automated. After that, I got interested in Python and GAN and made different things by myself as a result of playing around.
How did you start programming?
I started studying Python along with a book I bought, called ゼロから作るDeep Learning (Deep Learning from Scratch.)
What advice would you give you people who want to start programming from zero experience?
Don’t overdo it; I think it’s important that you enjoy it and don’t end up hating programming.
What was the inspiration for AI Gahaku?
I like making people happy, and I like programming, so I wanted to use what I liked to make something fun and interesting.
Did you make AI Gahaku on your own?
At the final stages of testing I asked my twitter followers for help, but everything else I did on my own.
How did you build your model? Can you describe how AI Gahaku works and its reference image data?
I can’t speak about it in detail, but I use a pix2pix base model, and a dataset containing a variety of old paintings as reference.
What was the most fun and the most challenging parts of building an AI model?
I think the most appealing part of it is that new methods are coming out one after another, and anybody can easily read research papers even if they don’t belong to a university. In terms of challenges, it takes a long time to see results in your work, and when things don’t go well there’s nobody nearby I can ask for help.
You mentioned that you would like to use a variety of big data to increase the output of AI Gahaku. In what kind of way do you want to increase its output?
I’m still not a GAN specialist so I’d like to study more into it. The model I’m using is a model that crashes easily, so I’d like to have a plan to improve that part of it.
Are there any other AI projects you would like to work on outside of AI Gahaku?
In general I like image processing. I’d really like to make something similar to AI Gahaku, something interactive that anyone can try out easily!
What do you enjoy about image processing and GANs?
This overlaps with the question about fun and challenge, but I like that anybody can access new AI/programming methods when they’re released, and it’s easy to see results in your work even for non-specialists like myself.
You can keep up to date with Sato’s forays into programming and AI by following them on Twitter (in Japanese).
For more machine learning interviews, guides, and news, check out the related resources below and don’t forget to subscribe to our newsletter.