Adobe’s AI Is Capable of Learning Painting Styles and Reproducing Arts in under a Minute

AI seems to be catching up when it comes to reproducing artwork. A recently published paper on by University of Maryland and Adobe Research details about how an AI can learn to reproduce handpainted canvases in under a minute. The paper is titled “LPaintB: Learning to Paint from Self-Supervision” and details an artificial intelligence model that is capable of reproducing hand-painted canvases. That’s not all, the AI can be set to follow famous painters like Leonardo da Vinci and Vincent Van Gogh’s style.

Researchers have used self-supervised learning to create AI model. In this technique, unlabelled data is used alongside labeled data and this is something that will help improve its learning accuracy. The AI then models the system’s action states mathematically. After supplying the final state, the AI model learned to paint reference design whilst adhering to certain artistic style.

“With the development of non-photorealistic rendering techniques, including stroke-based rendering and painterly rendering, specially-designed, or hand-engineered methods can increasingly simulate the painting process by applying heuristics,” wrote the co-authors.

The AI framework is capable of executing painting actions with controlled parameters like stroke size, color, and positional information. As far as training data is concerned, the team drew pictures from reference images and set the samples to a fixed size. A PC powered by a 16-core processor and an Nvidia GTX 1080 was used to reproduce a 1000 x 800 image with 20,000 strokes in less than a minute.

Meanwhile, the researchers warned that the training model is as efficient as the data it is being fed. Apparently, their method is based on basic painting environment. However, in the future one can build a model-based reinforcement learning framework which can be used in a painting simulator. On a related note, Adobe recently announced the Fresco artwork app for iPad with AI-assisted live brush feature.

[via VentureBeat]