Human preference feedback has significantly improved the performance of large language models but remains rarely explored in low-light image enhancement. Existing approaches typically rely on objective metrics (e.g., PSNR, FID), often resulting in enhanced images that lack aesthetic refinement and local controllability. To overcome these shortcomings, we introduce an aesthetic-guided diffusion framework tailored for illumination enhancement under low-light conditions. Specifically, we first introduce an aesthetic-aware evaluation model trained on paired human preference feedback, enabling optimization of diffusion models to align with human aesthetic judgments. Secondly, we design a prompt-driven brightness adjustment module to achieve precise local brightness and aesthetic enhancements on selected regions or objects. Comprehensive evaluations across diverse benchmarks confirm that our approach consistently surpasses leading methods, yielding superior perceptual quality while offering greater adaptability and controllability.