Search papers, labs, and topics across Lattice.
This paper presents Z-Image Turbo++, a novel 2-step image generation model distilled from an 8-step teacher model, addressing the challenges of limited model capacity and task difficulty in few-step generation. The authors introduce three key design innovations: Distribution-Aligned Adversarial Learning, Step-Decoupled Parameterization, and End-to-End Training with Iterative Regularization, which collectively enhance the quality of 2-step generation. Experimental results demonstrate that Z-Image Turbo++ significantly narrows the quality gap with 8-step generation, showcasing the effectiveness of tailored distillation strategies in improving efficiency without sacrificing output fidelity.
Z-Image Turbo++ achieves high-quality 2-step image generation that rivals 8-step models, transforming the landscape of diffusion distillation.
Few-step diffusion distillation has become increasingly mature for 4-8-step generation, yet pushing further to 2 steps remains challenging. In this work, we introduce Z-Image Turbo++, a high-quality 2-step image generation model distilled from the 8-step Z-Image Turbo teacher. Our method addresses the central bottlenecks of increased task difficulty and limited model capacity in 2-step generation through three simple but effective design choices tailored to this regime. First, we propose Distribution-Aligned Adversarial Learning, which uses teacher-generated images rather than external real images as real samples for GAN training, providing a more attainable and informative adversarial target. Second, we adopt Step-Decoupled Parameterization, assigning independent model parameters to the two denoising steps to better match their distinct capacity demands. Third, we perform End-to-End Training with Iterative Regularization, allowing the first step to receive gradients from final image quality while preserving a meaningful intermediate generation through an explicit step-1 loss. Together, these designs substantially narrow the quality gap between 2-step and 8-step generation in both qualitative and quantitative evaluations, highlighting the potential of carefully tailored distillation strategies for improving the quality-efficiency trade-off in few-step generation.
