With the growing complexity of deep learning methods adopted in practical\napplications, there is an increasing and stringent need to explain and\ninterpret the decisions of such methods. In this work, we focus on explainable\nAI and propose a novel generic and model-agnostic framework for synthesizing\ninput exemplars that maximize a desired response from a machine learning model.\nTo this end, we use a generative model, which acts as a prior for generating\ndata, and traverse its latent space using a novel evolutionary strategy with\nmomentum updates. Our framework is generic because (i) it can employ any\nunderlying generator, e.g. Variational Auto-Encoders (VAEs) or Generative\nAdversarial Networks (GANs), and (ii) it can be applied to any input data, e.g.\nimages, text samples or tabular data. Since we use a zero-order optimization\nmethod, our framework is model-agnostic, in the sense that the machine learning\nmodel that we aim to explain is a black-box. We stress out that our novel\nframework does not require access or knowledge of the internal structure or the\ntraining data of the black-box model. We conduct experiments with two\ngenerative models, VAEs and GANs, and synthesize exemplars for various data\nformats, image, text and tabular, demonstrating that our framework is generic.\nWe also employ our prototype synthetization framework on various black-box\nmodels, for which we only know the input and the output formats, showing that\nit is model-agnostic. Moreover, we compare our framework (available at\nhttps://github.com/antoniobarbalau/exemplar) with a model-dependent approach\nbased on gradient descent, proving that our framework obtains equally-good\nexemplars in a shorter computational time.\n