This paper introduces a method to analyze and control Generative Adversarial Networks (GANs) by identifying interpretable controls for image synthesis. The authors use Principal Component Analysis (PCA) to identify important latent directions in the GAN's latent space or feature space. They demonstrate that a wide range of interpretable controls can be defined by layer-wise perturbations along these principal directions. The method is applied to both StyleGAN and BigGAN models, showing that BigGAN can be controlled with layer-wise inputs in a StyleGAN-like manner. The paper also explores the properties of the identified controls, such as their ability to control geometric configuration, viewpoint, and appearance. The authors provide a user interface for interactive exploration of these controls and compare their results to random directions and supervised methods, showing that their approach can achieve similar effects with fewer entanglements. The paper highlights the potential for future research in analyzing and extending the capabilities of existing GANs.This paper introduces a method to analyze and control Generative Adversarial Networks (GANs) by identifying interpretable controls for image synthesis. The authors use Principal Component Analysis (PCA) to identify important latent directions in the GAN's latent space or feature space. They demonstrate that a wide range of interpretable controls can be defined by layer-wise perturbations along these principal directions. The method is applied to both StyleGAN and BigGAN models, showing that BigGAN can be controlled with layer-wise inputs in a StyleGAN-like manner. The paper also explores the properties of the identified controls, such as their ability to control geometric configuration, viewpoint, and appearance. The authors provide a user interface for interactive exploration of these controls and compare their results to random directions and supervised methods, showing that their approach can achieve similar effects with fewer entanglements. The paper highlights the potential for future research in analyzing and extending the capabilities of existing GANs.