This project presents CycleGAN-Turbo and pix2pix-turbo, which adapt single-step diffusion models like SD-Turbo for image-to-image translation tasks in both paired and unpaired settings. It uses adversarial learning to leverage pre-trained diffusion model knowledge while achieving fast inference (0.29 seconds on A6000 for 512x512 images). The models outperform existing GAN-based and diffusion-based methods for sketch-to-photo and edge-to-image generation.