A general-purpose framework for accurate and efficient diffusion inversion via noise embedding optimization, enabling few-step inference and robust performance on novel samples.