Abstarct: One of the modern approaches in image steganography is leveraging deep learning technology. In these methods, convolutional neural networks are employed, which have the ability to extract complex features and use these features as patterns for embedding hidden messages in images. Older methods faced challenges such as simple structures, limited capacity, low retrieval accuracy, imbalance between retrieval accuracy and stego-image quality, gradient vanishing issues, and insufficient capacity. In this research, a frxamework named VidaGAN, baxsed on deep learning, was developed. This network consists of three main components: an encoder, a decoder, and a critic, which partially addresses the challenges of previous methods. VidaGAN, utilizing generative adversarial networks, enables the optimization of the visual quality of generated images. This method achieved a steganographic capacity of 3.9 bits per pixel on the DIV2K dataset. Evaluation of the results using the StegExpose steganalysis tool showed an auROC value of 0.6, indicating acceptable security for the VidaGAN architecture. However, this method still faces limitations such as a small receptive field, limited capacity, and dependence on small datasets. To overcome these shortcomings and enhance performance, a more advanced approach baxsed on transformers has been proposed. This new approach replaces the downstream laxyers of VidaGAN with vision transformers to improve the analysis and understanding of complex image patterns. Transformers, with their ability to process long-range relationships in data, enhance the network’s learning capacity and expand its receptive field. Additionally, in the proposed second architecture, CoordConv blocks are used in the encoder and decoder sections, which enhance the network’s spatial awareness and improve the quality of stego-images. This new method, named VidaFormer, is designed to increase steganographic capacity, improve message retrieval accuracy, and enhance the quality of stego-images. VidaFormer, by integrating transformer architecture with advanced design features, mitigates VidaGAN’s limitations and achieves a steganographic capacity of 4.89 bits per pixel. This innovative method not only addresses existing challenges but also opens new horizons in the field of image steganography by providing an advanced frxamework.