This paper presents a diversity-based topology optimization framework for designing soft robotic grippers. The approach combines generative design and topology optimization (TO) to create a wide range of high-performance gripper designs. Compositional pattern-producing networks (CPPNs) seed diverse initial material distributions, which are then fine-tuned using TO. The focus is on vacuum-driven multi-material grippers, and the method is demonstrated to generate grasping modes such as pinching and scooping without explicit prompting. Extensive automated experimentation with 3D-printed multi-material grippers shows that optimized designs exceed the grasp strength of comparable commercial designs. The evaluation covers 15,170 grasps, assessing grip strength, durability, and robustness. The combination of fine-grained generative design, diversity-based processes, high-fidelity simulation, and automated experimental evaluation represents a new paradigm for bespoke soft gripper design, applicable across various domains, tasks, and environments. The main contributions include a framework for generative TO, a method for multi-material pneumatic soft robots, and the generation of diverse and high-performing soft gripper designs. The paper also discusses related work, the multi-material pressure-driven topology optimization formulation, and the CAD geometry generation process. Experimental results show that the optimized designs exhibit significant variability in morphology and performance, with some designs outperforming commercial grippers in terms of grasp strength, durability, and robustness.This paper presents a diversity-based topology optimization framework for designing soft robotic grippers. The approach combines generative design and topology optimization (TO) to create a wide range of high-performance gripper designs. Compositional pattern-producing networks (CPPNs) seed diverse initial material distributions, which are then fine-tuned using TO. The focus is on vacuum-driven multi-material grippers, and the method is demonstrated to generate grasping modes such as pinching and scooping without explicit prompting. Extensive automated experimentation with 3D-printed multi-material grippers shows that optimized designs exceed the grasp strength of comparable commercial designs. The evaluation covers 15,170 grasps, assessing grip strength, durability, and robustness. The combination of fine-grained generative design, diversity-based processes, high-fidelity simulation, and automated experimental evaluation represents a new paradigm for bespoke soft gripper design, applicable across various domains, tasks, and environments. The main contributions include a framework for generative TO, a method for multi-material pneumatic soft robots, and the generation of diverse and high-performing soft gripper designs. The paper also discusses related work, the multi-material pressure-driven topology optimization formulation, and the CAD geometry generation process. Experimental results show that the optimized designs exhibit significant variability in morphology and performance, with some designs outperforming commercial grippers in terms of grasp strength, durability, and robustness.