Producing of granular media with irregular shapes is one of the significant challenges in different fields related to granular particles. Representing inaccurate and simplified particles can result in wrong predictions. In this paper, a new statistical and optimization-based algorithm is presented by which 3D particles with complex shapes are generated on a random container. Stochastic conditional sampling along with an optimization method are combined such that particle with realistically and controllable shapes are generated. Depending on the initial shape of the container, a pack of spheres is generated. Then, an iterative optimization-based conditional sampling is proposed by which other complex shape descriptors are considered and can help to avoid artifacts and unrealistic shapes. The proposed method is applied to two examples wherein the particles reproduction and controllability of algorithm is demonstrated. The performance of the algorithm, in terms of shape reproduction, is also quantified using various utilized statistical measures. (C) 2019 Elsevier B.V. All rights reserved.