Large Multimodal Models have proven to be remarkably adept at comprehending tasks involving broad vision and language. However, these models frequently face difficulties when handling complex scene understandings and narratives because of the limited supported input resolution (e.g., 448 x 448) and the incomplete description of the training image-text combination.