High-precision geological models in smart mines typically feature massive data volumes and highly complex topological structures. Consequently, traditional geometric slicing methods struggle to meet millisecond-level interactive demands, while voxel-based approaches suffer from precision loss when delineating orebody boundaries. To address these issues, we propose a lightweight, real-time slicing algorithm tailored for mine visualization. By leveraging the parallel computing capabilities of GPUs, we introduce a pixel-masking slicing strategy that transforms complex geometric intersections into pixel-level retention judgments, effectively decoupling the computational load from the model’s geometric complexity. Furthermore, an attribute-indexed texture mechanism is developed. Through dual-layer “geometry–texture–attribute” mapping, critical geological data—such as lithology and ore grade—are baked directly into the UV space. This overcomes the traditional limitations of render-based slicing, enabling the real-time querying of stratum attributes. Experimental results demonstrate that for complex geological models containing up to 10 million facets, the single-slice response time remains strictly between 1.2 and 1.6 ms, with dynamic interactive frame rates sustained at approximately 60 frames per second. Notably, this performance exhibits no fluctuation despite exponential increases in mining area and geological body scale. The generated cross sections display crisp, anti-aliased boundaries and seamlessly support real-time interactive queries for ore grade, lithology, and reserves. Demonstrating both high efficiency and high fidelity, this approach resolves the bottlenecks of real-time spatial analysis in complex geological environments, providing robust visualization support for building smart mine digital twins and predicting deep-seated mineral deposits.