<p>This paper proposes a novel and scalable cryptographic framework for image encryption based on Chaotic Neural Networks (ChNNs). The key contribution lies in a hybrid architecture that combines a Chaotic Neural Network Long Short-Term Memory (ChNN-LSTM) model with a Chaotic Neural Network Multi-Layer Perceptron (ChNN-MLP). The ChNN-LSTM, trained on the Lorenz chaotic system, generates trainable chaotic features, which are then used to train the ChNN-MLP for predicting new chaotic sequences. These sequences, which lack explicit mathematical formulas, enable the generation of encryption keys and S-Boxes. To investigate the security and performance of the proposed cryptosystem, comprehensive statistical analyses—including entropy, NPCR, UACI, and correlation tests were conducted. The results demonstrate strong security, high randomness, and robustness against known cryptographic attacks. This makes the proposed method not only efficient but also practical for real-world applications such as secure communications, medical image protection, and biometric data encryption.</p>