The concept of auto mimesis, where one sees an image of itself in its own creation becomes more and more relevant now in the age of AI. More so as GANs (Generative adversarial networks) have really revolutionised autonomic learning, where the machine is pitted against itself and keeps learning and optimising till it can beat itself at its own game. While in many situations a clear goal like victory, classification, solution etc can clearly define how the machine is performing, but in other areas like art it becomes a subjective truth.

There are a couple of motivations here 1) the pursuit of learning about and furthering painting by exploring the various painting styles of master painters 2) how are the ai models are making their decisions while conjuring these images based on various instructions using artifical intelligence methods public image datasets like imagenet. Bearing in mind that neither the model nor the dataset has sufficient information to entail a precision study of the topic to make an informed opinion on the same. Neither does it intend to be reduction or generalisation of the works of the master painters. It is just a casual exploration in the interest of studying aesthetics through generative painting, by coaxing a machine to reveal its thinking process.

The deviations between painter's styles and ai model's approach on constructing these images are exposed by forcing the algorithm to render one of the most iconic painting "Monalisa" in various artist styles as it understands, so that it gives us some comparative basis for reading the deviations. Although various ai methods exist where transfer learning is used to apply one artist style over the other, this approach is different using variational auto encoders, where the image is constructed bit by bit, transforming the input attributes to map onto the attributes and features and relationships that the model has learnt itself. (See VQGAN and CLIP for more info). Collect on Objkt.com