Currently, diffusion architecture generative models have achieved significant success in areas such as image generation and image semantic inference. However, despite being text-driven models trained with extensive visual semantic data, their potential for image enhancement remains largely untapped. In this paper, we leverage a no-reference Image Quality Assessment (IQA) model to guide the diffusion model in enhancing image quality. Specifically, we independently trained four distinct no-reference IQA models focusing on brightness, contrast, colorfulness, and sharpness to direct the diffusion model in improving images from different aspects. Experimental results demonstrate that IQA models from different domains can enhance various aspects of image quality, although the effectiveness of these enhancements can vary significantly across different images.