In late 2016, StackGAN was the first generative text to image algorithm to show photorealistic results conditioned on the textual input. For many practitioners this work felt as being ahead of its time by one year. A few recent papers improve upon this, one of which is AttnGAN. I have used the latter to develop a novel real-time speech-to-image visual poetry machine called "Pictures of Jap Girls in Synthesis" (https://www.youtube.com/watch?v=lSFoQcqhbfU&list=PLZV8sKAO3LzZ4dxF57JbSiUj4CTqVBHd1). I will do a lightning review and introduction generative text to image methods using deep learning, while using the visual poetry machine for a live demo.