Our study presents a novel approach to sequence-to-sequence prediction for the image captioning task. We named this system as Bangla Chitra Net, and for this purpose we have used Transformer which receives sequentialized raw images as input. An image encoder is included in this transformer-based system which is based on convolution neural networks (CNNs), from the input image it extracts visual features which are based on regions. Also, recurrent neural networks (RNNs)-based caption decoder that uses the visual features and attention to construct the output caption words. Majority of the previous work has shown the effectiveness of intramodal interactions through co-attention, while most of the contemporary intramodal interaction techniques rely on self-attention. These approaches do have a limitation, though, in that the sequence must be processed sequentially. Some researchers have used the Transformer model to create captions from photos using English datasets in order to get around this limitation. But none of them used the transformer concept to produce efficient Bengali captions. As a result we utilized a transformer based efficient image captioning model. Extensive trials show that the suggested model is effective, and on the Flickr8k & BAN-Cap dataset, we outperform the traditional “CNN+Transformer” approaches.