Z-Image-Turbo-SDA is a highly efficient LoKr (Low-Rank Kronecker Product) adapter designed to rescue the "Diversity Collapse" problem in few-step distilled Flow Matching / Diffusion models.

By applying our novel Semantic Directional Alignment (SDA) Loss, this LoKr adapter recovers 70.2% of the original teacher model's compositional diversity (LPIPS) while perfectly maintaining the extreme sharpness and fast inference speed of the 8-step baseline model.