By using multimodal text embeddings extracted using encoders, an enhanced semantic relationship between the text input and the generated image is achieved. The GAN which begins with a crude, low-resolution image, and then improves it over multiple iterations to come up with a final image.
