Achieving a systematic understanding of embryonic development is a central goal of developmental biology. Digital embryo approaches provide an emerging framework for integrating multimodal developmental data toward digital reconstruction and predictive modeling of embryonic development. In this review, a bibliometric analysis was conducted to characterize research trends and emerging topics in digital embryo research and related fields from 2010 to March 2026. A total of 9 770 articles were retrieved from the Web of Science, of which 8,622 articles were retained after topic-based filtering. Keyword analysis identified scRNA-seq as the predominant technology, with spatial transcriptomics and multi-omics technologies emerging as major research themes. Early embryonic development was a major biological focus, while blastoids, gastruloids, and retinal organoids were prominent in vitro models. Based on bibliometric results, we synthesize recent advances across five essential pillars: single-cell multi-omics atlases, spatial omics landscapes, in vitro developmental models, bioinformatics analysis, and artificial intelligence virtual cells (AIVCs), highlighting their distinct contributions to digital embryo research. Single-cell omics technologies characterize embryonic states at molecular resolution and reveal both coordinated and decoupled dynamics across omics layers. Spatial omics preserve cellular location and tissue context, extending developmental atlases from molecular-state annotation toward spatiotemporal reconstruction. In vitro models provide experimentally tractable platforms for direct observation, controlled perturbation, and mechanistic validation, while supplying complementary data for digital embryo modeling. Multi-omics integration, spatial mapping and alignment, and three-dimensional reconstruction constitute key computational tasks in digital embryo construction. AIVCs learn representations of cell states and developmental dynamics, thereby supporting the prediction of perturbation responses and cell fates. We group current methods for in silico perturbation prediction into three recurring strategies: latent-state representation learning, biological network-informed modeling, and distributional transport, involving frameworks such as (variational) autoencoders, graph neural networks, optimal transport, and flow matching. Recent AIVC approaches are increasingly being applied to developmental biology, demonstrating the potential to model developmental trajectories, reconstruct spatiotemporal dynamics, and predict perturbation-induced changes. Overall, digital embryo research is transitioning from spatiotemporal multi-omics atlas construction toward predictive virtual-cell modeling. At the data level, major challenges include incomplete spatiotemporal coverage of developmental atlases, limited multimodal datasets, and a scarcity of perturbation data. AIVCs remain constrained by limited interpretability and out-of-distribution generalization, as well as a lack of standardized benchmarking and systematic experimental validation. We propose AIVC modeling of preimplantation embryos as a plausible near-term objective, with the potential to advance developmental biology and reproductive medicine.