The sequential Point-of-Interest (POI) recommendation plays a vital role in today’s location-based social network platforms. While numerous studies have proposed various promising solutions for improving POI recommendations, the field still faces key challenges: (1) Data sparsity is very high, and the number of user visit records is limited; (2) Most existing models ignore the power of different modalities to capture deep user-level preferences. In response to these issues, we propose a novel Cross-Modal Sequential POI Recommender with Lightweight Hybrid Fusion Strategy (CMPLH), which is designed to enhance the performance and efficiency of sequential POI recommendations. By incorporating the Geo-Local Sensitive Hashing (Geo-LSH) attention mechanism, our model effectively fuses multiple modalities to capture dynamic user preference features from multi-modal historical check-ins, while reducing computational overhead. Furthermore, by leveraging a hybrid fusion strategy, CMPLH effectively integrates users’ historical POI check-ins, categorical sequences, geographical information, and user-generated reviews. Extensive experimental results confirm that CMPLH surpasses existing state-of-the-art approaches, demonstrating the advantages of combining multiple modalities and hybrid fusion strategies in enhancing sequential POI recommendation systems.