Visible-infrared person re-identification focuses on matching identities across visible and infrared images, with the key challenge being extracting discriminative features across modalities. To tackle this, we propose an Automatic Visual-Language Aligning Network (AVLA). It leverages a multi-modal pre-trained model to automatically generate text descriptions for pedestrians during training, and incorporates an Image-Text Dynamic Attention (ITDA) module to categorize and fuse features by modality. Additionally, we design weighted distribution matching (WDM) to align visual and textual embeddings globally. Extensive experiments on public datasets demonstrate that our method outperforms existing approaches.