Color constancy, the ability to maintain consistent color perception under varying illumination, remains a central challenge in computer vision. This is particularly true in complex, mixed-lighting scenarios where traditional algorithms falter and even modern deep learning methods can introduce visual artifacts. To address these issues, we propose WaveDiff-CC, a novel framework that effectively combines three key technologies. Our method first leverages a wavelet transform to decompose the image into a multi-scale domain. We then introduce a conditional diffusion model to generate robust, data-driven priors to guide the color correction process. Finally, a dynamic multi-head attention architecture adaptively fuses these multi-scale features under the guidance of the priors, enabling seamless and global color reconstruction. Extensive experiments show that WaveDiff-CC achieves state-of-the-art performance in terms of accuracy and visual quality on challenging datasets, providing a powerful and effective solution for color constancy in real-world scenarios.