Inflatable devices such as soft robotic systems, biomedical balloons, and deployable architectural components must follow controlled deformation trajectories during pressure-driven actuation to safely navigate obstacles, satisfy geometric constraints, and avoid mechanical failure before reaching a final deployed state. In these systems, intermediate configurations are often as critical as the final shape, since transient contact, geometric interference, or stress concentration can compromise functionality or structural integrity. This work introduces a latent space generative framework for the inverse design of inflatable structures with prescribed, mechanically safe deformation trajectories. Deformation trajectories are specified using low-dimensional signals that describe how key geometric quantities evolve during inflation, including changes in overall dimensions and geometry-based safety descriptors extracted from the deforming shape. High-throughput finite element simulations of pressure-actuated structures generate large datasets of inflation-induced deformation trajectories across variations in undeformed geometry, spatially varying material properties, and boundary conditions. These trajectory descriptors, together with the corresponding structural configurations and mechanical responses, are embedded in a latent space and used to train the generative model and associated surrogate models that replace repeated forward simulations during design exploration. The trained framework enables the generation of undeformed inflatable configurations that evolve along target trajectories under pressure loading while respecting mechanical constraints. Numerical examples with increasingly complex target deformation paths demonstrate the framework’s ability to capture path-dependent behavior and uncover coupled relationships between undeformed geometry, material distribution, and pressure-driven shape evolution.