The inverse design of acoustic metamaterials presents significant challenges due to the complex, non-intuitive mapping between desired acoustic properties and corresponding geometries. This project explores the application of diffusion models to solve this problem, leveraging their proven capability in handling one-to-many mappings. A dataset of 2D metamaterial geometries and their associated acoustic dispersion properties, including bandgap widths and locations, was generated using Gaussian field geometries and Finite Element Analysis. The diffusion model, guided by prompts for bandgap properties, learns to generate geometries through a stochastic denoising process. A U-Net-inspired architecture was employed to capture the relationship between geometry and acoustic properties, with prompt injection at the latent space level to bias the denoising process. Preliminary results demonstrate a robust and extensible geometry generation method along with the diffusion model's ability to learn the denoising process from this rich dataset. Moving forwards, this research will refine the model architecture and augment the training dataset to establish greater performance and hopefully demonstrate robust capabilities in inverse design of acoustic metamaterials, paving the way for potential applications in acoustic filters, dampeners, and waveguides.