Low-light image enhancement is a challenging task aimed at improving the visibility of low-light images, and the results obtained from previous methods still have shortcomings in terms of detail recovery and overall quality. Recently, diffusion models have shown impressive ability in generating realistic and detailed images through a multi-step denoising refinement process, which motivates us to introduce them into low-light image enhancement for higher quality results. However, we found in experiments that using a diffusion model to enhance low-light images leads to color distortion in the restored results. To address this issue, we propose a model called Two Diffusion Streams (TDS). It employs a dual-stream architecture, where one stream is used to initially enhance the low-light image and reconstruct its structure, while the other stream utilizes the color map for color correction. The experimental results on existing benchmark datasets demonstrate that our model not only resolves the color distortion issue but also produces higher-quality results with fewer noise and artifacts. Additionally, it exhibits stronger generalization capability across different scenes.