AI-based multimedia data generation systems require cloud computing capabilities for training large-scale model with over billions of parameters but can experience high latency due to the remote nature of the cloud. Edge servers can offer lightweight model training, fine tuning and real-time inference services, which are less computationally and storage intensive. Therefore, this Recommendation identifies a reference framework and requirements of AI-based multimedia data generation systems using core cloud and edge cloud. The use cases are also provided in an appendix.