Abstract: At present, the average number of cameras in coal mines exceeds 500, making continuous manual monitoring impossible and large screens insufficient for comprehensive display. Reviewing videos afterward cannot promptly identify potential safety hazards, making it difficult to prevent accidents. Therefore, Artificial Intelligence (AI)-based video monitoring in coal mines has become an inevitable choice under the principle of "no operation without video". Conventional AI models have problems such as poor generalization and difficulty in recognizing abnormal events. Large AI models can solve these problems and should possess functions including data collection and indexing, large-model pre-training, fine-tuning and deployment, and model iteration, while supporting multimodal data including images, videos, text, and audio. General-purpose large AI models are trained using large-scale data from the mining industry to form large AI models for mining. Large AI models for mining have advantages such as powerful functions, good generalization, strong versatility, and high reliability. Large AI models for mining should have a scenario generation function that allows users to use a small amount of labeled data to automatically generate scenario models through workflows, meeting the application needs of mine production, safety, geological surveying, transportation and sales, coal preparation, operations, and management. Based on the safety production requirements of coal mines, the functional, interface, data, software and hardware platform, and deployment requirements of large AI models for mining are proposed to standardize their planning, design, engineering construction, operation and management, and maintenance, and to promote the application of AI in coal mines.