James She, Carmen Ng, Wadia Sheng: Keep Running – AI Paintings Of Evolving Horse Portraits

Keep Running is a collection of human and machine generated horse paintings using the generative adversarial network technology. The artworks are produced during the lockdown period in the Middle East due to the Covid-19. Horses are significant symbols in the Middle East region culturally and historically, representing strength, endurance and persistence. These are the spirits that keep us running during the difficult times, even when we are facing many physical constraints in daily life. Nowadays, many AI artworks are either photo-realistic or very abstract with distorted faces, fragmented figures and a combination of unknown objects. What technically unique in our work is to produce a series of AI-assisted paintings that shows distinguishable features and forms of horses, while creating aesthetic and even sentimental values in each horse portrait. Each of our art piece is presented in a 2×2 grid format that shows how an AI horse painting is evolved over the generation process. We believe that the machine learning process could unite human creativity with AI technology to produce a series of unique and aesthetic paintings – even these artworks were created in a large scale and mass production method that is never been possible before. With the artistic and technical novelty in our work, we also wish to pay a salute to the pioneers like Eadweard Muybridge and Andy Warhol, who first popularised the use of machines like Camera and Silk-Screener in art-making, redefining the meaning and expanding the horizon of art.

Many Generative Adversarial Networks (GAN) technologies can nicely generate photo-realistic images, or the other extreme of images with abstract hybrids with distorted face, fragmented figures or unknown combination of objects. To master a fine balance in between these two extremes, it is very challenging to generate a painting of a common subject in art history like a horse with complete features and figures along with aesthetic styles and even sentimental perception. We modified certain parameters (e.g., AdaIN gain and bias values) in our StyleGAN’s algorithm to train the machine using our collected images of horse paintings and photo-realistic images. Here what we achieved is the fine balance between these extremes, in order to produce a set of stylish and artistic paintings of horses with clear features and figures, and even sentiments. Finally, we just let the machine generate every painting on its own, and select some evolving moments that are aesthetically and interestingly being put together with its final result in a grid format. Each piece can be exhibited physically by a 60CMx60CM display, or online by an animated image with 1024×1024 pixels.

Some argued that Andy Warhol’s silk screening art pieces are lacking high art qualities and fine art values, as they were mass-produced by machine with the image clippings from newspaper or magazines. Warhol, however, happily called himself a machine, and even named his art studio as “The Factory” in the 60s, produced a large amount of silk-screening artworks under the help of a team of assistants. Although the image of the same subject was reprinted again and again, such repeated process and generation actually gave life to Warhol’s most known work – “Marilyn Diptych”. With the help from AI, we produced a set of similar repetition art of horse portraits using GAN technology with a massive data base of horse paintings and photo-realistic images. From the top left corner in each canvas, one can see how a stylish and artistic horse portrait was being generated and was evolving from pixels to a brush strokes stage by stage. With the same 2×2 grid presentation, one could be further amused by how the same repeated machine learning process could in fact produce different and unique horse paintings in different stages. Eadweard Muybridge and Andy Warhol have both proven that even with the use of machines, like camera and silk-screening machine, the uniqueness of a machine-created artwork won’t be diminished or sacrificed; same for our artwork, we want to emphasize and celebrate the use of machine and mass production through AI technology, while showing the possibility to create, explore and discover more unpredictable but aesthetic and interesting values in AI art making. This perfectly aligns to theme of “Untitled and Untied” in SIGGRAPH 2020 Asia.

Nowadays, many AI-generated artworks are either photo-realistic or very abstract with distorted faces, fragmented figures and a combination of unknown objects. What technically unique in our work is to produce a series of AI-assisted paintings that shows distinguishable features and forms of horses, while creating aesthetic and even sentimental values in each horse portrait. 

To train our AI generative algorithm, collecting the required number, type and quality of horse images from the local horse stables is the biggest challenge, especially during the COVID-19 lockdown period. The following video described our behind-the-scene story. 

http://cyphymedia.com/aiart/keeprunning//