Autonomous materials discovery with desired properties is one of the ultimate goals for materials science, and the current studies have been focusing mostly on high-throughput screening based on density functional theory calculations and forward inference of physical properties using machine learning. In this work, we developed and applied a generative deep learning algorithm to predict distinct stable crystal structures by optimizing the formation energy in the latent space. It is demonstrated that the optimization of physical properties can be integrated into the generative model as on-top screening or backward propagator. Such an implementation has been successfully applied on the binary Bi-Se system, showing that distinct crystal structures can be obtained covering the whole composition range, and the phases on the convex hull can be reproduced after the generated structures are fully relaxed to the equilibrium. Recently, our approach has been extended to multicomponent systems trained using more than 52,000 compounds in the Materials Project database, where novel compositions can be generated covering most elements in the periodic table. This paves the way for multi-objective optimization and finally inverse design of materials with optimal properties.