What is neural style transfer and how does it use deep CNN features?

Neural style transfer combines content and style representations using CNN feature activations.

CNNs encode hierarchical features: early layers capture texture, later layers capture semantic content. Style transfer optimizes an image to match content features of one image and style statistics (Gram matrices) of another. This separates content and style representations in deep feature space.