The Unintended Benefit of Mapping a GAN's Latent Space

While trying to improve the quality and fidelity of AI-generated images, a group of researchers from China and Australia have inadvertently discovered a method to interactively control the latent space of a Generative Adversarial Network (GAN) – the mysterious calculative matrix behind the new wave of image synthesis techniques that are set to revolutionize movies, gaming, and social media, and many other sectors in entertainment and research. Their discovery, a by-product of the project's central goal, allows a user to arbitrarily and interactively explore a GAN's latent space with a mouse, as if scrubbing through a video, or leafing through a book. An excerpt from the researchers' accompanying video (see embed at end of article for many more examples). Note that the user is manipulating the transformations with a'grab' cursor (top left). The method uses'heat maps' to indicate which areas of an image should be improved as the GAN runs through the same dataset thousands (or hundreds of thousands) of times.