In the contemporary digital era, spreading false information via social media and Internet platforms poses a significant challenge. Not only the data has become increasingly multimodal, lack of sufficient labeled data due to prohibitive labeling costs and data privacy issues poses an additional challenge. An approach such as meta-learning can help address this issue. This work showcases meta-learning with scarcely labeled data for the detection of multimodal fake news and provides a performance comparison of the meta-learning approach with the fully supervised learning approach. We leverage the power of Model-Agnostic Meta-Learning (MAML) and Fast Context Adaption via Meta-Learning (CAML) in this work. The results are encouraging as these meta-learning models achieve heightened accuracy and speed even with insufficient labeled data for multimodal fake news detection.