Convolutional Neural Networks

Until very recently, artwork was reserved for creatives with a defined idea of self-expression, applying a specific lens to their world to convey thoughts and feelings through pieces that would take hours–and sometimes years–to create.

The evolution of AI and convolutional neural networks (also known as ConvNet) has pushed the boundaries further than ever, allowing everyone to create astonishing artworks in just a few minutes. Machine learning means you can turn photos to paintings online, prototype a piece of art over one hundred times faster, and make photos look like pop art with technologically advanced tools.

What Are Convolutional Neural Networks?

ConvNet has a secret power, which all lies in style transfer processes. Style transfer is the technique we use to recompose an image into any style you've chosen. If you want to make a photo look like an oil painting, NightCafe can make it happen!

From a user perspective, you follow a few simple steps:

  • Upload your source image
  • Pick a style that inspires you
  • Choose your resolution
  • Generate your artwork

Behind the scenes, AI is replicating an artistic style to reimagine an existing graphic with ConvNet, the deep learning technique that makes it possible. There are countless ConvNet applications, including image classification and analysis, which powers things like facial recognition cameras.

The key is training convolutional neural networks to interpret content and style images while grasping the shapes, colours, perspectives, and shades that define them. Without deep diving into the fascinating complexity of AI, ConvNet has multiple convolution layers, separating the elements on your graphic and extracting them piece by piece before rebuilding them.

How Can AI Convolutional Neural Networks Create Artwork?

For many people, the idea that AI can develop authentic pieces of art is challenging to comprehend. Still, it's impossible to overstate how quickly AI advancements have accelerated. While people can understand and interpret art in a very different way from a machine, the capacity of AI means it can synthesise the third image, using your graphic as the input, and the second to dictate the rules of texture and style. 

Before deep learning stepped onto the stage, researchers had been playing with handcrafted ways to extract content from an image, but neural networks make easy work of this task. ConvNet learns as it goes, expanding those hidden layers underneath the neural net, which comprise non-linear optimisations, partial derivatives, equations, and correlation estimators we don't need to know about to create each piece of art!

The basics, though, are straightforward–ConvNet extracts the content and style and merges them to create something new and utterly unique.

How Long Does It Take to Make Artwork With ConvNet?

Here's a more relatable question: how long does this super-intelligent AI technique require to rebuild a complete piece of art, drawing on its two input images as a starting point? That depends on the size of the images and how detailed the origin graphics are.

AI can transform small images with one colour channel in seconds; ConvNet could classify about 1,000 simple images in under 0.1 seconds. However, high-resolution images with multiple colours can take a few minutes each. Either way, it's an extraordinary development and a space where AI becomes a genuinely useful tool, with applications from medical imaging to simply building beautiful artworks you can enjoy forever.