Variational Autoencoders (VAEs) are powerful deep generative models widely used for representation learning, data generation, and anomaly detection. However, their performance is highly sensitive to hyperparameter choices, such as the dimensionality of the latent space, the weighting of the Kullback-Leibler (KL) divergence term, and network architecture specifics. Manual tuning of these hyperparameters is time-consuming and often suboptimal, while automated methods like Bayesian Optimization or Random Search can be computationally expensive, especially for complex VAE architectures and large datasets. This article proposes a novel meta-learning approach to automate the hyperparameter optimization (HPO) process for VAEs. By learning from the HPO experiences across a diverse collection of previous tasks (datasets), the meta-learner can predict promising hyperparameter configurations for new, unseen tasks, significantly accelerating the optimization process. We demonstrate the effectiveness of this approach by building a meta-dataset of VAE performance across various data characteristics and training a meta-model to recommend optimal hyperparameters. Our results show that the meta-learning framework can efficiently identify near-optimal VAE hyperparameters, leading to substantial computational savings while maintaining competitive model performance, thereby advancing the field of automated machine learning for generative models.