Synthetic simulations of planetary system formation can provide access to correlations between the properties of planets in the same system. Such correlations can, in return, be used to guide and prioritize observational campaigns aiming at discovering some types of planets, based for example on the presence and properties of outer giant planets. Nevertheless, these can be very demanding in term of computing power. To address this, we present a conditional generative model based on a transformer model (Alibert, Davoult & Marques, in review), trained on synthetic systems from the Bern planet formation model. This model can quickly generate realistic planetary systems, while taking into account stellar and disk properties such as metallicity, mass, and lifetime. The generated systems reproduce key features of the original simulations, including planet multiplicity, orbital spacing, and mass distribution. This new approach allows us to produce large scale statistics which can be used to predict the presence of possible unseen planets following observations and study further the properties of planets and the planetary systems including long-period giants.