GANs are an important part of the development of generative AI technologies. These models are capable of generating synthetic data from random noise input or with some directed inputs. Image processing has also greatly benefited from GANs, as image-to-image translation allows for transforming any visual aspect of an image into a desired one. Examples include image restoration, enhancement, and transformation. Image translation techniques are being used in many areas such as agricultural automation, self-driving cars, satellite image processing, photo editing, and even in critical applications like medical image processing. GANs utilize many deep learning techniques such as multi-layer perceptrons, CNNs, and transformers. The potential for automated plant disease diagnosis to increase agricultural yields and decrease pesticide use has drawn a lot of interest in recent years. Although plant disease identification using deep learning approaches has shown promise, these methods frequently suffer from overfitting, which impairs their ability to generalize findings to new data. Attempts have been made to use unpaired image translation techniques to convert photographs of healthy leaves to images of diseased ones. CycleGAN-based methods transmit latent details from their training data to the translated image, which limits the potential of using it to generate different images. However, they perform well on unpaired I2I translation. In this paper, we examined the research on plant leaf image-based decline categorization as well as the work on using GANs to enhance datasets of cucumber leaf pictures in order to improve classifier accuracy. Expand on our suggested Transformer Assisted CycleGAN (TACGAN) approach, which creates a variety of diseased cucumber images from healthy photos by utilizing a transformer-based GAN architecture. By employing a transformer-based methodology, our technique is able to concentrate on pertinent regions of the images and enhance the training dataset with diverse backgrounds, thereby increasing the adaptability of the image classifier that has been trained on enhanced images.