For this series, I am automating the generation and selection of abstract painting compositions and then painting them with a desktop robotic arm. The paintings are inspired by abstract expressionist paintings focusing on colors and mark-making.
In these paintings, the background was painted by hand and a robotic arm used to paint the responding brush strokes layered on top. Using a classifier to pick the best compositions, I have noticed the highest ranked compositions tend to align to and highlight the background curves, as well as create interesting color combinations.
Figure 1 – Desktop robotic arm painting brushtrokes to create the asbstract painting on the right.
This work is inspired by abstract expressionism painters who create large color field and mark made paintings that evoke experiences to the viewer. Often these paintings were created intuitively and spontaneously, by the painter. Automatic can mean “without conscious thought or intention” by a human and be considered as artistic genius, but it also means working without human control – like with a robot.
Here are a few of my most recent Automatic Abstract paintings:
Figure 2 – Recent “Automatic Abstracts” that have been created as daily paintings at my studio. Each is 9″ x 12″ on heavyweight paper
These paintings start with a washy, single color background that I hand paint with a large paint brush and water. This unique under-painting is the background for the robotic painting process. My software then generates and reviews thousands of potential designs to place on top of this underpainting.
Figure 3 – Me in my studio just having created the backgrounds for the starting point of this project.
I let the software create hundreds of thousands of digital painting compositions on top of the under-painted background, I then sort through them with a classifier trained on my preferences. The training is based on my artistic preferences – “Good” or “Bad”.
Often multiple good brushstroke sets are created for the same background – I then decide which is the best one to paint.
Figure 4 – Eight digital brush stroke sets that have been created and layered on top of a potential pre-painted background. The classification algorithm has sorted them based on my artistic preferences, it ranked the images on the left as low (not to paint) and the ones on the right as high (potential successful paintings). The ones on the left have more visual interest that responds to the background shape and color rather than hiding it (on the left).
Brush Stroke Constructions
The robotic arm is programmed to mimic how I apply paint to a canvas. I use one color at a time, my motion tends to be curved because my own arm that has pivot points – wrist, elbow, shoulder and waist. Because I do not reorient myself while painting a section, the brush strokes are parallel within a set. To create visual interest in the painting, and take advantage that a robot is good for patterns – which I’ve programmed it to do with a small brush. The robot also, like my painting process, cleans the paint brush between colors and waits for the paint to dry before continuing.
Variety: I am currently programming more styles of brush strokes to create, classify and paint.
Dataset: I am also using the final paintings to become their own dataset that an algorithm can create new paintings from using a GAN rather than hard coding the brush stroke sets. As well as using a GAN to decide if a painting is complete.
Feedback Loop: I am also looking to add vision into the system to create a feedback loop in the system and assess the painting as it is created.
I am excited to share that this work was shown at the Seymour Art Gallery in Deep Cove, Canada in July 2019 and won the Carolyn Badgley Emerging artist award. I also showed it at the O’Reilly Artificial Intelligence Conference in London UK in October 2019. It will also be part of the online gallery of the 2019 NeurIPS conference in Vancouver Canada. A few images of the installations are below.
Figure 7 – Far Left – Image from Seymour Art Gallery installation. Middle & Far Right – Images from the O’Reilly Artificial Intelligence Conference.