One of the biggest problems faced regarding images in various fields is the generation of meaningful description for the image i.e. caption for the image. The images are being used for numerous purposes with major on the web however they had to spend a great share of their time to generate a proper and accurate description for the image. This makes it very complex as the machine has to learn from the datasets and then describe the objects, activities and the places. The fact that humans can do it quite easily for small sets but fail when the number of images is more. This make it a rather interesting challenge for deep learning algorithms. The applied approach for the image caption generation would be based on long-short-term memory networks (LSTM) and recurrent neural networks (RNN). Such network model allows to select the next word of the sequence in a better manner. In this paper, Python is used to form this caption generating platform with the help of TensorFlow library which can easily generate the LSTM model for a given images. In this research work, machines are trained by deep learning approach. To improve the efficiency of the caption generation, the training has to be quite deep with more sample images. Additionally, detailed analysis is done on the improvement which can be brought to implement by including Beam Search in it. © 2019 Journal of Applied Science and Engineering. All rights reserved.