The new technique enables the generation of sharper, more lifelike 3D shapes—like these robotic bees—without the need to retrain or finetune a generative AI model.