A text-to-image model with support for native high-resolution (2K) image generation