Molar: Multimodal LLMs with Collaborative Filtering Alignment for Enhanced Sequential Recommendation