General adversarial networks (GANs) can be applied to synthesize photorealistic images with up to 1024×1024 pixels. However, GAN-based image generation is computationally expensive, which limits its applications (including at the edge). To address this problem, Sergei Belousov proposes MobileStyleGAN, a high-resolution image generation model that provides comparable quality to StyleGAN2 but with far less parameters and computational complexity.