Guidance from Keyword for Identifying Small Details in High-Resolution Images with Vision-Language Models