There are many classes of Petri nets for describing communicating systems. Some of these guarantee important properties, such astermination in the case of portnets. There are also many methods andtools available for their analysis and synthesis. However, when developing new methods, or benchmarking against existing ones, it is oftenhelpful to quickly generate large sets of random models satisfying certainproperties and user-defined rules.This paper presents a heuristic-driven method for synthetic generation ofrandom portnets based on refinement rules. The method considers threeuser-specified complexity parameters: the expected number input andoutput places, and the prevalence of non-determinism in the skeletonof the generated net. An implementation of this method is availableas an open-source Python tool. Experiments demonstrate the relationsbetween the three complexity parameters and investigate the boundariesof the proposed method.