The generation of crystal structures is a big problem for crystallography and crystal structure prediction [1]. Until now, an open question is the creation of random crystal structures in which the positions of some structural elements are already known. This problem arises when the positions of some atoms can be determined using experimental techniques. One of the important example is superconducting hydrides, where positions of heavy atoms can be determined using the X-ray, yet hydrogen positions remain unknown. Another example is molecular crystals and cocrystals of high energy materials, where positions of solvent molecules can also be unknown. Crystal structure generators that exist today do not allow adding and fixing known structural information (i.e., environment).In this work we propose a novel approach to the generation of crystal structures sets. It allows one to enter known parameters of a system, such as cell size and atomic positions, and to specify the structural information to be added: types of atoms or/and molecules, limitations in interatomic and intermolecular distances. Position density can also be varied and set manually. The generator determines whether the generated structure belongs to one or another cluster of previously generated structures, comparing structural information using the ALIGNN graph neural network, and thus provides the necessary structural diversity of the resulting set on the fly.We illustrate this method by applications to uranium, lanthanum and sulfur hydrides at different pressures and to CL_20 × N2O and CL_20 × CO2 molecular crystals at standard conditions. Generated structures were used as seeds in the evolutionary algorithm USPEX [2], where they were relaxed with VASP at given pressures. Using our method, we found stable phases of all researched systems, which correspond to experimental results from [3], [4], and [5] for superconducting hydrides and [6] for high energy solvates of CL_20, respectively.[1] A.R. Oganov, C.W. Glass, The Journal of Chem. Phys., V. 124, Is. 4, 2006[2] A.O. Lyakhov, A.R. Oganov, H.T. Stokes, Q. Zhu, Comp. Phys. Comm., V. 184, pp. 1172 – 1182, 2013[3] I.A. Kruglov, A.G. Kvashnin, A.F. Goncharov, A.R. Oganov, Sci. Adv., V. 4, I. 10, 2018[4] I.A. Kruglov, D.V. Semenok et. al, Phys. Rev. B 101, 024508, 2020[5] A.P. Drozdov, M.I. Eremets, , I.A. Troyan, V.Ksenofontov, S.I. Shylin, Nature, V. 525, pp. 73 – 76, 2015[6] J. Xu, S. Zheng, et. al, Chem. Comm. 2019, Is.7
