Accurate and early detection of plant diseases is crucial for maintaining agricultural productivity and crop quality. Guava, a prominent fruit crop, is particularly susceptible to various diseases that can significantly impact its yield. Manual identification methods are often inefficient, time-consuming, and dependent on expert analysis. To address these limitations, this study explores the use of advanced hybrid deep learning models for automated disease detection in guava leaves. The research utilizes an annotated dataset sourced from Kaggle, containing images of guava leaves affected by multiple disease types. Three hybrid architectures, DenseNet201 + GRU, Inception v3 + GRU, and InceptionResNetV2 + GRU, were implemented to classify diseases. These models effectively combine the spatial feature extraction capabilities of convolutional neural networks (CNNs) with the temporal sequence learning ability of GRUs. The DenseNet201 + GRU model achieved an accuracy of 99.47%. The Inception v3 + GRU model demonstrated improved performance with an accuracy of 99.68%. The best results were obtained with the InceptionResNetV2 + GRU model, which achieved an accuracy of 99.75%. The hybrid models leverage their ability to process intricate spatial and temporal patterns in disease-affected leaf images, setting a new benchmark for automated plant disease classification. This study highlights the potential of integrating CNNs with GRUs for precise and scalable disease detection, offering a practical tool for farmers and agricultural stakeholders. By providing accurate and reliable diagnostics, this work contributes to more efficient disease management and promotes sustainable agricultural practices.