A new study presents a zero-shot learning (ZSL) framework for maize cob phenotyping, enabling the extraction of geometric traits and estimation of yields in both laboratory and field settings without the need for model retraining.