Artificial intelligence in mental health has emerged as a potent tool to foster precision psychiatry, for example, by stratifying patient populations. A potential step forward would be mental health digital twins-the independent in-silico reconstruction of an individual person within their functional social and environmental systems that continuously incorporate all known and available subject parameters to predict patient trajectories including the outcomes of interventions. Generative artificial intelligence in the form of large language models demonstrated the ability to mimic human responses and integrate diverse sources of information that may foster the development of digital twins. We give a brief historical perspective on concepts and milestones of artificial intelligence in mental health and outline the current state of clinical decision support systems, monitoring and therapy applications based on artificial intelligence. We describe their integration in large behavioral models as a recently met precondition for digital twins and contrast this development with the magnificent hurdles that remain to truly realize clinical benefits of digital twins, from data quality and regulatory compliance to user engagement and public trust, for some of which we propose mitigation strategies here.