Some initial experiments using generative adversarial networks to transfer stylistic characteristics from one image to another.
Module Drawings mix geometrical tiling systems with various pencil style drawings. These pieces are an attempt to reduce the complexity of my initial tests with GANs. Although working with flowers produced some visually beguiling images, it was hard to grasp how the machine was interpreting and transferring styles. These new tests used primitive geometrical shapes as a means to make more meaningful observations from the process at hand.
The above two images show two very different styles rendered from two different data sets that hold samples of pencil line drawings. The three images below are generated from the same source image yet have also very different textures.
The output samples below are further explorations demonstrating varying graphical styles. What is revealing in this rather slow but meticulous process is that one is forced to observe images in a detailed manner. Personally, it trains the eye and makes me appreciate stylistic qualities on a more engaging level. What I find particularly interesting with the visual results is that these stylistic characteristics break the mathematical precise line of traditional generative art. Geometric abstract art has always guided my work when it comes to creating generative patterns and I can appreciate its pristine expression. A lot of great work continues to be seen with this movement and especially so within the generative disciplines.
With this new process, I feel I'm opening another door for exploration. One that takes root in the visual language of geometric abstraction yet extends it with a more organic set of aesthetics.