Allows multimodal models to pass image files to themselves for analysis, including images found online and downloaded by other tools.