For building robust and generalizable reactive MLIP training set, preparation of optimized initial structures is needed. This module can generate reasonable 3D structures frome SMILES strings and re-optimized with GFN-xTB method to retain consistency with following processes.