With the accelerating integration of artificial intelligence (AI) and the real economy, video analysis capabilities based on AI algorithms have become a very important requirement for many vertical applications in video surveillance. With the popularity of artificial intelligence research, artificial intelligence algorithms have been implemented in intelligent video surveillance (IVS) system. However, these AI models face these challenges when implemented in IVS systems: (1) the generalization ability of AI models is weak, and the algorithm performance of AI models is poor in different video surveillance scenarios; (2) Data quality is crucial for the training and implementation of AI small models. There is a lack of high-quality labeled data in the IVS system. To better promote the application of IVS, ensure the application effect, and promote the research of IVS technology, many companies have launched IVS systems with high functional integration based on large-scale pre-trained generalized artificial intelligence models to meet the diverse needs of the IVS. This has led to a highly complex, heterogeneous, and fragile ecosystem. This Recommendation aims to address the issue of research and development in the industry but lacking a unified standard framework. This Recommendation provides an overview of the model generalization system in intelligent video surveillance, and specifies the requirements and framework for model generalization system.