Try a graphical editor based on neural network

All we can draw simple shapes or sketchy people, cars and so on. But the real talent of the artist is endowed with a few. Oh and if it comes to photoshop, then to personal qualities is added the need to understand the intricacies of the program. But you’ve probably heard that high technology to facilitate our lives? Now, the moment has arrived: thanks to a new neural network right now, any of you without any special knowledge and skills can paint a pretty good picture.

The development meet is not the last people from the world of science, namely researchers from the Massachusetts Institute of Technology (MIT) and experts from IBM called the IBM Watson AI Lab. To create the drawings thanks to the special generative neural network GAN (Generative Adversarial Networks).

What do I need to do? In fact, nothing special. You have a set of elements: clouds, trees, grass, buildings and so on. As well as the steps that with these items you can do to add to the picture or to remove them. It looks very interesting. You just select the object and action, and then sketch the cursor to a specific area on the photo. Everything else makes the neural network.


Naturally, the neural network works not flawlessly. For example, I would like to note that best it is to create\remove the miraculous objects like trees and grass. They look as natural as possible, while the facades of the buildings often look pretty ugly. Especially if you try to remove them or replace with something else.

This is the turret sticking out from behind either trees, or grass for me to clean up and failed

And here is a reverse example. Originally pictured in the lower part was grass. It turned out clean, and the right part to fit the bushes. Looks very good.

Rounded roof, added a round dome on the tower, removed the grass and replaced some of the trees on the buildings

Yes, the neural network does not always cope with the task, but the first computer was not much computing power, occupying several rooms. This is only a demonstration of future technologies, with due attention will develop into something new and interesting.

In the meantime, you can try yourself as an artist on the official website of the project. This is a very addictive process. In addition, if you are interested in the internal structure of neural networks — you can also watch the video available below.