Dielectric capacitors are crucial for advanced electronics due to their high power densities, but their energy density needs improvement. High-entropy strategy is effective but discovering new systems within a high-dimensional composition space is challenging. This study uses a generative learning approach to accelerate the discovery of high-entropy dielectrics. By encoding-decoding latent space regularities, inverse design screens promising combinations. Five targeted experiments yielded a Bi(Mg<sub>0.5</sub>Ti<sub>0.5</sub>)O<sub>3</sub>-based high-entropy dielectric film with an energy density of 156 J cm⁻³ at 5104 kV cm⁻¹, surpassing the pristine film by over eightfold. This method drastically reduces experimental cycles and extends to other multicomponent material systems.
