Optimizing Few-Step Generation with Adaptive Matching Distillation
Researchers have developed Adaptive Matching Distillation (AMD), a new framework to improve the stability and performance of few-step generative models. AMD addresses issues in "Forbidden Zones" where existing distillation methods struggle by using reward proxies to detect and escape these problematic areas. Experiments on image and video generation tasks, including SDXL and Wan2.1, show AMD enhances sample fidelity and training robustness, notably improving the HPSv2 score on SDXL. AI
IMPACT Enhances training robustness and sample fidelity for generative models, potentially leading to more efficient and higher-quality AI-generated content.