Scaling up GANs for Text-to-Image Synthesis