Virtual brain twins are personalized, generative, and adaptive brain models based on individual brain data, designed for scientific and clinical applications. These models aim to mechanistically explain and capture relevant data features to guide decision-making in diagnostics, prognosis, and therapy. The standard model for personalized whole-brain network models integrates various concepts and methods, emphasizing neural mass and neural field large-scale modeling, nonlinear dynamics, and network science. Personalization involves three stages: assembling cortical and subcortical areas in the subject-specific brain space, mapping connectivity and other parameters into the brain models, and estimating relevant parameters through model inversion using probabilistic machine learning. The article discusses the use of personalized whole-brain network models in healthy aging and five clinical diseases: epilepsy, Alzheimer's disease, multiple sclerosis, Parkinson's disease, and psychiatric disorders. Key challenges and future directions include addressing degeneracy, overfitting, and improving accuracy through higher-resolution and multiscale models. The authors highlight the potential of deep generative models and high-resolution simulations to enhance the predictive power and precision of virtual brain twins. In epilepsy, the virtual epileptic patient (VEP) model uses personalized brain models and machine learning to estimate epileptogenic networks and aid surgical strategies. For Alzheimer's disease, the model focuses on regional variability and the impact of amyloid-β and tau proteins. In multiple sclerosis, the model addresses slower conduction velocities and structural lesions. For Parkinson's disease, the model incorporates the accumulation of misfolded α-synuclein and degeneration of dopamine-producing neurons. Psychiatric disorders, such as schizophrenia, are modeled using spatial masks to represent changes in neurotransmission and neuromodulation. The article concludes by emphasizing the importance of addressing challenges like degeneracy and overfitting, and the potential of advanced models for precise and accurate predictions in brain disorders.Virtual brain twins are personalized, generative, and adaptive brain models based on individual brain data, designed for scientific and clinical applications. These models aim to mechanistically explain and capture relevant data features to guide decision-making in diagnostics, prognosis, and therapy. The standard model for personalized whole-brain network models integrates various concepts and methods, emphasizing neural mass and neural field large-scale modeling, nonlinear dynamics, and network science. Personalization involves three stages: assembling cortical and subcortical areas in the subject-specific brain space, mapping connectivity and other parameters into the brain models, and estimating relevant parameters through model inversion using probabilistic machine learning. The article discusses the use of personalized whole-brain network models in healthy aging and five clinical diseases: epilepsy, Alzheimer's disease, multiple sclerosis, Parkinson's disease, and psychiatric disorders. Key challenges and future directions include addressing degeneracy, overfitting, and improving accuracy through higher-resolution and multiscale models. The authors highlight the potential of deep generative models and high-resolution simulations to enhance the predictive power and precision of virtual brain twins. In epilepsy, the virtual epileptic patient (VEP) model uses personalized brain models and machine learning to estimate epileptogenic networks and aid surgical strategies. For Alzheimer's disease, the model focuses on regional variability and the impact of amyloid-β and tau proteins. In multiple sclerosis, the model addresses slower conduction velocities and structural lesions. For Parkinson's disease, the model incorporates the accumulation of misfolded α-synuclein and degeneration of dopamine-producing neurons. Psychiatric disorders, such as schizophrenia, are modeled using spatial masks to represent changes in neurotransmission and neuromodulation. The article concludes by emphasizing the importance of addressing challenges like degeneracy and overfitting, and the potential of advanced models for precise and accurate predictions in brain disorders.