In the bone marrow, immature lymphocytes proliferate rapidly and abnormally in acute lymphoblastic leukemia (ALL), a potentially fatal hematologic malignancy that disrupts the normal production of blood cells. This disease accounts for around 25% of pediatric malignancies, making it very common among youngsters. Improving treatment results requires an accurate and timely assessment. Even though AI holds promise for promoting early detection, existing AI-based diagnostic models often face drawbacks like high processing requirements, overfitting from unbalanced data, and poor generalizability—all of which limit their practical clinical effectiveness. A MultiNet-Based Ensemble model that combines ResNet50 and VGG16 is presented in this work to enhance the categorization of ALL from peripheral blood smear pictures. EfficientNet-B0 (96.32% accuracy), VGG16 (97.39% accuracy), and ResNet50 (98.31% accuracy) were assessed separately. Advanced data augmentation, class-weighted loss, and optimization using Adadelta, SGD, RMSprop, and Adam are all included into the ensemble. Across the Benign, Early, Pre, and Pro subtypes, hyperparameter tweaking produced an accuracy of 99.85% and an F1-score of 0.9985. The proposed method significantly enhances ALL diagnostics due to its remarkable effectiveness, computational efficiency, and clinical integrability; nonetheless, external validation is required to address potential overfitting and guarantee practical application.