The significance of automatic image captioning systems, which generate precise and contextually appropriate descriptions, has increased due to social media’s rapid development. In this paper, an efficient model for generating image captions is presented, which combines the usage of Long Short-Term Memory (LSTM) networks for sequence generation with Convolutional Neural Networks (CNNs) for visual feature extraction. Through the use of sophisticated attention mechanisms and optimization on datasets specific to social media, our method outperforms present techniques in generating captions that are both more accurate and contextually appropriate. By means of extensive research, we demonstrate the accuracy of our model, attaining higher BLEU scores than with conventional techniques. Our improved training procedure and the addition of attention mechanisms, which boost the ability of the model to produce captions that closely resemble human descriptions, are responsible for this performance improvement. Our research has significant practical consequences, especially for social media platforms where precise automatic captioning can improve user engagement, enhance accessibility for users with visual impairments, and support content moderation. The application of our methodology is addressed comprehensively, along with key concerns and prospects for automated image caption generation.