The classification of violent scenes in cartoon videos is a complex task due to the stylized and exaggerated nature of cartoon visuals. To address this challenge, we propose a deep learning-based framework that integrates Convolutional Neural Networks (CNNs) with residual blocks and Long-Short-Term Memory (LSTM) units to effectively capture both spatial and temporal features from video frames. A dataset of 2,400 images was used, divided into five categories: Explosion, Blood, Fight, Gunshot, and Normal (non-violent). To address class imbalance, the dataset was augmented using Deep Convolutional Generative Adversarial Networks (DCGANs). The proposed architecture combines convolutional layers, residual connections, and LSTM components to model the intricate spatial details and temporal dynamics of cartoon violence. This framework demonstrates significant potential for applications such as content moderation and automated media analysis. Experimental results highlight the effectiveness of the model, achieving an accuracy of 97.59%, with sensitivity and specificity of 94.5% and 95.8%, respectively. These findings underscore the robustness of the proposed architecture in accurately distinguishing violent and non-violent scenes in cartoon videos, offering a reliable solution for this challenging task.