Inverse design in materials, which consists of input as desired properties and output as fine-tuned materials that satisfy given criteria, is a state-of-the-art method to obtain desired materials efficiently. However, it is still difficult to contrive robust and accurate inverse materials design platform to fit user’s requirements thoroughly. In this work, we observed that a previous platform which integrates a genetic algorithm with deep learning can be a robust tool for inverse design using generation of metal-organic frameworks (MOFs) for selective xenon adsorption as a case study. By using our platform, we obtained two MOF candidates that shows exceptional xenon/krypton selectivity over the current record in computational simulation. Furthermore, we demonstrate that our platform can work with complicated conditions such as multiple properties and a range of property values by facile modification in the cost function of genetic algorithm. With this result, we can expect that our flexible platform can use as universal method to generate finely tuned MOFs that fit the specific desires of users.