It is common knowledge on the web and in digital libraries that results with an image result in far more engagement than those without. In contrast to textual features, our understanding of the visual cues is fragmented at best, and visual information is typically ignored in search, recommendation, and engagement analysis. We extend the most used news recommendation dataset (MIND) with the lead image of the news items, annotate each image using News Value Theory factors, and experiment with different LLM modalities to detect these factors. Our research provides a practical way to use currently ignored visual features as additional handles for search and recommendation in mixed media collections.