This week I explored the possibility of a machine-learning algorithm analyzing a poster. This is particularly significant to me because this displays the possibility that an AI can understand the separate parts of a visual sequence, and if it can understand them, then it can use that understanding to construct something similar.

The difference between this and the traditional way of generative networks generating posters after being trained on a dataset consisting of posters, is that the traditional method is more akin to one of us holding a paintbrush and trying to construct the idea of a poster by simply painting shapes, rather than actually placing typography and images to create an effective composition.

--

--