Computer Vision Art

Jeremy Kraybill

028749_0001_08, (2018)


028749_0001_08 was my first robot & AI-generated painting and received 6th place in the 2018 robotic art competition. It was a technical proof of concept for a series of paintings I am currently working on. I used several GAN and discriminator neural networks, trained on late 1800’s expressionist portraits, to generate a source 64x64 pixel thumbnail. Hundreds of thousands of candidate images were generated, and scored using networks trained on personal taste parameters. The final image was then translated into a set of paint mixing and brush stroke instructions for an ABB IRB 120 industrial robot arm. The painting is gouache on watercolor paper. It used approximately 4,000 brush strokes and was painted in a single 3-hour session with human intervention only to replenish the 5 primary paint colors and fix last-minute bugs. Special thanks to Loren Adams and the team at the University of Melbourne School of Design for their support and assistance in making this project a reality.

028749_0001_08 028749_0001_08 028749_0001_08