somehow midjourney “knows” lots of context so it can make art that captures what you’re asking. i suspect this will only get better over time with multimodal models.