Steganography conceals information inside another object, so information is not immediately apparent in the examination. Different from encryption, where a ciphertext is an encrypted object, steganography aims to hide information and the fact that a piece of information is concealed. In the case of image steganography, the secret is embedded into a cover image so that the cover image does not differ substantially, and the secret can be extracted without too many alterations. This paper proposes a method to embed an image into another image using U-Net architecture with MobileNet Convolutional Neural Network. The proposed method is trained and tested on a readily available image dataset, CIFAR10, StanfordCars, and STL10. The result shows that the proposed method can embed and extract images from CIFAR10, StanfordCars, and STL10 datasets with average PSNR of 28.165 dB, 28.539 dB, and 28.226 dB for steganographic images; 27.075 dB, 29.759 dB, and 28.767 dB for extracted images.