Lawrence Lee, acclaimed Contemporary Southwestern artist, and Bård Ionson, coder with a keen eye and an artistic bent, have begun a fascinating journey of collaboration that melds paintings with modern artificial intelligence technology.
Lee’s “magic people” inhabit a separate reality that seems to be of another world–a multidimensional one. By combining one of the original human creative outlets, painting, and advanced math with new technologies, a multitude of mages, seers, shamans and sages has been born.
This is a creation built on a lifetime of Lee’s creations at the easel. Bård used over 250 Lee paintings to train a machine learning / artificial intelligence system and used his own creative skills to perfect the output of the software and to curate results. Lee then worked to identify the best of the generated images and used his digital painting skills and immense creativity to improve on what Bård produced. With each iteration, new possibilities were revealed, and the pair are excited by the prospect of further development, incorporating new technologies as they become available and following the lead of some of the images produced thus far into new, previously unimagined areas.
Bård is now taking the improved results to teach the AI model all over again.
The goal was to expand the creative palette like a hallucinatory dream. Controlled by Bård with training selections, the AI produced a googleplex of possible random outcomes. Lawrence and Bård have hand-selected the best of each production run from the machine, and Lawrence has worked to unearth these new shamans and the landscapes they inhabit by enhancing them further and augmenting their otherworldly qualities in an attempt to better understand their roots and to release their powers.
Ionson and Lee will be releasing the series of images created, called Convergence, as weekly package drops of three still images and one video on SuperRare. Find them starting on June 23. The winning bidder will have the option to redeem an exclusive 1 of 1 , 10″ X 10″ inch print valued at $250 created by Lawrence in his Arizona atelier. Coupon code will expire after two months if not redeemed.
The mystical Shamans of Lawrence Lee are famous in the genre of American Southwest Art. Lawrence has painted thousands of quality paintings of these Shaman. Which is just one of the things that attracted me to work with him.
I got to know Lawrence through our work together on establishing an industry-wide convention for secondary sale royalties. For months a group of artists met virtually on Discord to discuss and plan ways to promote artist resale rights. He has also given art career advice to many of us based on his half-century of experience in the competitive art market.
I approached Lawrence about a collaboration knowing that he had a lifetime of artwork to build on. I suspected that we just might be able to make something special by combining my somewhat unique artificial intelligence art techniques. One thing that makes artificial intelligence art work right now is a large library of content. One needs thousands of images to train a GAN (Generative Adversarial Network) on how to make copies of an oeuvre.
He was very enthusiastic, and within an hour we agreed to work together.
Parts of his work really drew me into thinking this could work. I have an affinity for southwest art from growing up in Denver Colorado. He has lots of work. Often his work is square, which is ideal for AI art as code is built for 1:1 ratio work. His work has an otherworldly feel that I thought would lend itself to the artifacts that GAN introduces. And I really wanted to experiment with subject matter that was original and unique.
Start of an Idea
This was partly inspired by Robbie Barrat’s collaboration with Ronan Barrot, a French painter, called Infinite Skulls. Ronan has painted hundreds of small square skull paintings over a lifetime, and Robbie trained a machine learning / artificial intelligence GAN with them. Robbie curated what the GAN generated. Then Ronan would paint over them to fix them or he would paint new ones based on what was produced.
Lawrence liked the concept and we decided to move forward. He sent me 200 or so images of his shamanistic works. Unfortunately, many photos of his early Shaman paintings were lost. I was able to pull a number of them from various places on the internet. And with some manipulation, I was able to make many variations of those by altering them slightly–or sometimes drastically–using AI style transfer techniques.
In the process, I built about five different models and ended up with three that were sufficient. One key is that the GAN was not able to build good looking torsos and faces at the same time. One model ended up being trained on half-length subjects and the other on head shots. And then we have the original third model that makes surrealistic blobs with eyes and Shaman as viewed on a bad trip.
As I experimented with each model’s output I sent Lawrence a large batch of images. Some were random and some were curated choices by me. He says I have overloaded him with a year’s worth of ideas and images. He began finding the best ones and then experimenting with what to change to make them look like a Shaman should.
It was difficult because the first batches were fairly poor. He provided input on what needed to be corrected in the training and I took some of his initial experiments on fixing them and added them back into the training set. I removed thousands of images that were “corrupting” the training. Many of these bad images were introduced by my attempts to augment the training set. So now I directed the training on the faces by providing more headshots. We iterated on this.
The faces were just full of chins and mouths or just blobs. But Lawrence started enhancing them.
Then Lawrence had the idea of finding the best parts from multiple images and putting them together. Or perhaps I thought he said something along those lines. But in the middle of the night trying to sleep I think why not make a separate model for heads and build on top of an existing open-source model of faces. By cropping the heads I could build a separate model just for heads. I also include headshots and self portraits of Lawrence Lee.
In the meantime Lawrence created amazing pieces by augmenting the output of the GAN using Procreate and an iPad. Sometimes he invented an ancient god from just an abstract blob.
Once I was able to produce better faces I sent him a large batch of those. I had also been producing latent morphing videos commonly seen with GAN based art. He found that by stopping the video in the middle of a transition between faces he could find some good faces to build on.
We have ideas for improving the models. The next attempt might be to use high resolution images of just the faces including the new digital works. And also making sure they are all in a face forward position.
Part of the fun I have in making artificial intelligence art is the strange things it can produce when experimenting and adding things to the model just to throw it off a bit. This has been a rewarding and joy producing project to work on.
With his skills in manipulating color and shadow he has pulled out the best features of these beings. I am so excited to present these to the world. They are so much better than I could ever dream of being able to create.
Lawrence is one of the small number of fine artists in the space of tokenized art because he is always looking for ways to expand and learn new things. He has been a professional artist for over 40 years. He is one of the original adopters of non-fungible tokens on SuperRare for his digital art. In addition he has created computer art from early in the personal computer age.
Bård Ionson is an artist who is a relative beginner with art but has spent a career working with computers and programming. He is now creating digital art and video art using oscilloscopes, scanners and artificial intelligence technologies.