Counterfactual explanations, which explore “what if” scenarios post-hoc, have recently attracted significant attention for clarifying the predictions of complex machine learning models. However, its application in recommender systems remains relatively unexplored, particularly in generating explanations that are both accurate and diverse. This paper introduces RICE, a novel historical interaction-based counterfactual explanation method tailored for multimodal recommendation systems. Unlike existing approaches that often fall short in providing effective and varied insights into recommendation decisions, RICE produces counterfactual explanations that not only have short distances but also offer a range of alternatives. Extensive experiments on multiple benchmark datasets and multimodal recommendation models validate the effectiveness of RICE.