Neural Style

What if van Gogh had the imagination of Dali, or Dali the vision of van Gogh?
A new technique, developed by researchers in Germany, combines the style of one image with the content of another:
A Neural Algorithm of Artistic Style

(pyCaffe implementation here)

(TensorFlow implementation here)

The original method is quite slow but can be turbo-charged by training neural nets specifically for individual styles:

Deep Art

Fast Style Transfer

(TensorFlow implementation here)

Now a single network can be optimised for multiple styles:

Supercharging Style Transfer

Real-Time Style Transfer for iOS

CycleGANs to Create Computer-Generated Art

Ever wanted to be a pointillist, but never had the patience? Help is here.

Where the deepdream algorithm seeks to maximize the activations of particular layers of a deep neural network, this method starts with noise and gradually creates an image which matches the activations of lower layers by the style image, and a higher layer by the content image.