Seismic inversion methods, like most methods for solving an inverse problem, aim to derive a solution—a 3D model characterizing petrophysical properties from abundantly monitored seismic response data. This inverted model targets the reproduction of the observed data – real seismic. Inherent to this class of inversion problems is their ill-posed nature, lacking a unique solution. Deterministic, Bayesian, Full Waveform, and Geostatistical Seismic inversion methods offer a unique solution, often accompanied by limited uncertainty measures, which are quite constrained around the obtained result. In this study, a new approach is presented based on the Niching Genetic Algorithms method, to allow a diverse set of optimal solutions in a multimodal space, thus overcoming the major limitation of seismic inversion methods, which is to quantify the uncertainty of the final models. In the new seismic inversion approach proposed in this study, in each iteration of the optimization process, niches of petrophysical properties models are calculated with a Machine Learning Clustering method based on a distance measure of the similarity (closeness or remoteness) between models. For the set of niches, the evolutionary process (Niching Genetic Algorithm) will produce different solutions, but all close to the observable real seismic data giving rise to a multi-solution seismic inversion methodology. The new proposal is illustrated in a synthetic case study to perform risk assessment at the early stages of exploration.