Abstract： The short-term evolution of macrophysical and microphysical cloud properties serves as a crucial indicator of future weather evolution. Compared with numerical models and optical flow-based nowcasting methods, Deep Learning (DL) approaches better meet the requirements of cloud-condition nowcasting in terms of timeliness, spatial resolution, and accuracy. Although DL-based research on cloud properties remains less extensive than that aimed at radar reflectivity and precipitation nowcasting, a growing variety of model architectures and data fusion strategies has emerged to address specific task objectives, necessitating a systematic classification and comprehensive review of existing studies. Driven by the increasing demand for cloud nowcasting across weather modification, aviation safety, solar power forecasting, and disaster prevention and mitigation, this paper first evaluates the suitability of major DL architectures for cloud-related tasks. Recent domestic and international research advances targeting both cloud macrophysical (e.g., cloud top height, cloud cover, and cloud optical depth) and microphysical properties (e.g., liquid water content, ice water content, and droplet number concentration) are reviewed separately, covering model architectures, input data sources, and task-specific methodologies. This is followed by an in-depth examination of different technical approaches for integrating DL models with cloud physics parameterization schemes in numerical models, which offers a complementary pathway to improving the explicit forecast of cloud liquid water content, ice water content, and related variables. Furthermore, DL-based cloud-related operational systems implemented or under development in major meteorological services worldwide are summarized. Based on an analysis of the key technical challenges in short-term nowcasting of cloud macrophysical and microphysical variables, this review identifies promising directions for future development. For cloud macrophysical nowcasting, future efforts should focus on incorporating more complete physical mechanisms governing cloud formation and dissipation, as well as advancing probabilistic forecast frameworks based on generative models. For cloud microphysical nowcasting, which remains at an early stage, potential directions include developing time-series prediction models leveraging high-quality cloud microphysical retrieval products as training data, exploiting macro-to-micro cross-scale transfer learning by pretraining on abundant cloud macrophysical observations, and fine-tuning with scarce microphysical retrievals, and improving NWP-driven microphysical nowcasting through DL-microphysics scheme coupling.