This repository implements the Image Analogies paper using deep learning. It extracts feature maps from VGG16 (a convolutional neural network) and uses PatchMatch for efficient approximate nearest-neighbor patch matching between images. The system blends features from a source/target pair to apply analogous transformations to a new image, combining CNN features with Markov Random Field constraints for coherent image synthesis. It runs on TensorFlow or Theano and was adapted from the Keras neural style transfer example.