Research questionHow can reference-guided image generation control distinct attributes of multiple objects in complex scenes?A single reference image can contain several objects whose attributes need to be used independently. Global conditioning and a single identifier token can blend these cues, making it difficult to control each object's contribution in a new image.