Stable diffusion (SD) models have proven highly effective in generating quality images when trained on large amounts of natural data. However, their applicability to medical images needs to be further assessed. In this study, we explore the potential of generating abdominal Computed Tomography (CT) slices by fine-tuning a text-conditioned SD pipeline, and we examine the adaptation of the pipeline for grayscale images, which is expected to perform better in the medical imaging domain. Furthermore, we propose a novel evaluation method that considers the Hounsfield Units equivalence of different structures to assess the quality of the generated images. We show promising results that could have practical applications, such as helping to train medical professionals and improving the training of deep learning models. However, additional experiments are necessary to enhance the proposed technique for medical imaging in clinical settings. Furthermore, it is necessary to continue researching on validation metrics and assessment methods in the medical field.