This paper focuses on the effective utilization of multimodal information in multimodal recommendation systems. Previous research in this field can be broadly categorized into two approaches: explicit and implicit utilization of multimodal information. Explicit utilization directly incorporates modality features extracted from items into the network computation. However, this approach may suffer from suboptimal performance when the available multimodal information is limited. On the other hand, implicit utilization learns the representations of users and items by constructing auxiliary structures, such as multimodal graphs. These structures help implicitly recall relevant items from semantic perspective. However, this approach belongs to the local utilization of multimodal features due to the Top-k sampling and is not well suited for handling large amounts of modal data. To address these limitations, we propose a novel model called DAMORE (Dual-Aligned Model for MultimOdal REcommendation), which combines explicit and implicit utilization to help model achieve SOTA performance in scenarios with sparse multimodal features and large data scales. In implicit alignment, to provide a boarder perspective for local utilization, we introduce a novel user-user graph and combine it with item-item graph from previous works. In explicit alignment, we employ a self-supervised learning method to align item representations with multimodal features. Furthermore, previous approaches allow multimodal features to be updated during training, leading to semantic shift of original modal information. Based on this phenomenon, we introduce Modal Feature Persistence (MFP) to ensure feature stability during training. We conduct extensive experiments on three public datasets, and our model achieves state-of-the-art (SOTA) results. The code is accessible at this url .