The dissemination of hateful content in social media provides a significant obstacle to societal unity and individual mental health. This paper analyses the complicated challenges inherent in the identification of hate speech in the video formats, with a specific focus on native languages. In this paper, multimodal framework that employs advanced machine learning architectures, such as LSTM, BiLSTM, and Transformer models with MuRIL, XLM-RoBERTa, and IndicBERT is proposed to analyze visual elements. A majority voting mechanism is employed with the convolutional neural network (CNN) for visual analysis, which embolden the accuracy of hate speech in regional languages. The results highlight the effectiveness of the pre trained models along with the multimodal hate content detection.