This paper introduces a novel zero-reference low-light enhancement framework that does not require low-light data or illumination-specific hyperparameters. The key idea is to develop an illumination-invariant prior derived from the Kubelka–Munk theory of light transfer, which serves as an intermediary between normal and low-light images. The framework is trained solely using normal light images and employs a pretrained generative diffusion model, Stable Diffusion, to map the illumination-invariant prior back to images during testing. To address the challenge of restoring illumination-invariant information to images, the authors introduce a bypass decoder to handle detail distortion and a lightweight version of the model for practical applications. The framework is evaluated on various low-light datasets and demonstrates superior performance in terms of interpretability, robustness, and efficiency. The physical quadruple prior, which captures the essence of imaging under diverse lighting conditions, is a novel learnable illumination-invariant prior derived from light transfer theory. The framework also includes a prior-to-image mapping system that utilizes the prior as a condition to control the diffusion model. The results show that the proposed framework achieves favorable subjective and objective performance across diverse datasets, making it a promising solution for low-light image enhancement.This paper introduces a novel zero-reference low-light enhancement framework that does not require low-light data or illumination-specific hyperparameters. The key idea is to develop an illumination-invariant prior derived from the Kubelka–Munk theory of light transfer, which serves as an intermediary between normal and low-light images. The framework is trained solely using normal light images and employs a pretrained generative diffusion model, Stable Diffusion, to map the illumination-invariant prior back to images during testing. To address the challenge of restoring illumination-invariant information to images, the authors introduce a bypass decoder to handle detail distortion and a lightweight version of the model for practical applications. The framework is evaluated on various low-light datasets and demonstrates superior performance in terms of interpretability, robustness, and efficiency. The physical quadruple prior, which captures the essence of imaging under diverse lighting conditions, is a novel learnable illumination-invariant prior derived from light transfer theory. The framework also includes a prior-to-image mapping system that utilizes the prior as a condition to control the diffusion model. The results show that the proposed framework achieves favorable subjective and objective performance across diverse datasets, making it a promising solution for low-light image enhancement.