Deep Learning: A Tool for Creating Artistic Pastiches by Combining Different Styles
Google has created a new specialized neural network that allows users to blend different painting styles and create new pastiches in real-time.
The network can combine styles from different artists like Monet, Doré, and Picasso, resulting in new images with a unique combination of their distinct features. For example, you can create an image of your cat in the style of Van Gogh.
Previous systems for “style transfer” relied on neural networks trained on large databases of an artist’s work or analyzed a single artwork in detail. However, these systems required complex calculations to interpret a new image in the style of, for example, Claude Monet.
The latest work from Google Brain simplifies style transfer significantly. This system allows not only real-time image editing in different styles but also blending of multiple styles simultaneously.
Instead of repeatedly recreating the target image (your cat) until it resembles the source image (the painting), this new system takes a more advanced approach. Instead of learning the appearance of a specific painting, the new style transfer network learns the common style shared by multiple paintings from the same artist. The system analyzes several paintings by Monet and their effects on various examples to identify underlying similarities. These similarities can include specific color palettes, brushstroke styles, and other elements.
You might think you already have this feature on your phone. Apps like Prisma and others offer a similar effect. However, these apps use separate specialized networks for each art style. You select the style you want, and the app’s servers perform the calculations on the Van Gogh system and output the result. In this case, a single, super-efficient neural network handles everything, recognizing and combining dozens of styles based on lower-level features.
This advancement could significantly impact the fields of art and creativity.
