Identity Crisis XXVI - AI Portrait

A Generative Adversarial Network was first trained on 2,000+ images of abstract paintings: learning colors, textures and other raw characteristics of paint.

Next by transfer learning on top of the existing model with a new dataset containing 4000+ images of faces and portraits... 10,0000+ steps later, recognizable characteristics were revealed in the infinite array of outputs.

The final output is an animation flowing between generated images commonly referred to as a “machine hallucination”.