A widespread diffusion of genotyping is still a challenge in certain breeding programs, primarily due to prohibitive costs and the lack of appropriate genotyping resources for some plant species. Consequently, there is a growing interest in alternative approaches that can provide more easily accessible means of phenotype prediction. Phenomic prediction (PP) captures epigenetic, environmental, and dynamic influences that cannot be fully explained by genomic data alone. Recently, PP methods have been employed to complement genomic prediction (GP) approaches, addressing a key limitation of linear genomic prediction models in predicting complex traits, namely, their inability to account for genotype-by-environment interactions. On the other hand, multi-omics technologies have led to novel strategies for PP and GP, and the integration of complementary omics layers, by providing a more comprehensive view of the molecular mechanisms underlying phenotypic variation, has emerged as a promising approach to enhance prediction accuracy. The use of integrative models, capable of combining heterogeneous variables, allows to weight each level according to its predictive power and to capture nonlinear interactions between levels. Datasets can be managed by machine learning and deep learning tools, which are becoming crucial to extract meaningful patterns from these vast and heterogeneous resources.