Generating outputs (like images) based on learned query vectors that extract and guide specific information from input conditions.