_{Some of my favorite creations from the weekend.}

Dear friends,

in yesterday's shared Instagram post, I told you, that I spent the whole weekend exploring GanBreeder, "a collaborative art tool for discovering images using BigGAN."

Using the web app, you can mix or "breed" neural networks with each other into evocative, otherworldly imagery.

Source

What the GAN?

A GAN (Generative adversarial network) is a set of two competing neural networks, one is called the generator and the other the discriminator.

[...] The generator is creating new images that it passes to the discriminator. It does so in the hopes that they [...] will be deemed authentic, even though they are fake. The goal of the generator is to [...] lie without being caught. The goal of the discriminator is to identify images coming from the generator as fake.

Source

_{More GanBreeder creatures, I created.}

While GANs can be taught to create "worlds eerily similar to our own in any domain: images, music, speech, prose", GanBreeder focusses on creating new images using neural network models from BigGAN, a GAN setup and run by Google’s DeepMind division.

BigGAN uses images from ImageNet, which is a database of thousands of categories created from thousands of images.

Source

_{Another set of my GAN creations.}

Using Google's computational power and the large ImageNet database, the images created by BigGAN have a higher resolution (and 'accuracy'), than similar GAN's.

Down the Rabbit Hole

When you first visit GanBreeder, you have the options to either choose a randomly generated image, pick from a category or create an image using several genes (meaning categories) instead.

In this example, I've chosen four random categories. When you click on "create images", you will be provided 6 different images, that use a mix of these categories.

Choosing the most interesting image gives you the opportunity to further explore the mix by creating so called "children", by "crossbreeding" with a certain image or by editing the "genes".

If you like a certain image, you can create an account and press the heart next to the creation to bookmark it and share it with others. You can also download up to 5 images in high resolution, see the "Image Lineage" and share your creations on twitter. There is also a place, where you can see the creations of other users including a "Featured" gallery.

You can also pick an interesting image from these categories to explore it further. If you like a particular image a lot, there is even a website called GANvas Studio, that allows you to have your creations printed as posters or on canvas.

Art?

Since Christie’s sold a portrait, that had been generated by a GAN, for $432,000 last year, there is no doubt about the artistic abilities of these AI-algorithms. Like AI researcher Janelle Shane says:

You could illustrate a story this way, or make a hauntingly beautiful movie set. [...] It all depends on the data set you collect, and the outputs you choose. And that, I think, is where algorithms like BigGAN are going to change human art–not by replacing human artists, but by becoming a powerful new collaborative tool.

Source

What do you think?

Please share your thoughts in the comments!

---

Links

Here are some links for you to further explore the topic:

• My GanBreeder Profile
• GANvas Studio
• A Beginner's Guide to Generative Adversarial Networks (GANs)
• A Google intern built the AI behind these shockingly good fake images
• How three French Students used borrowed Code to put the first AI Portrait in Christie’s
• Wikipedia: Generative adversarial network