Images are an integral part of aphasia therapy (a speech disorder), and image collection is a time- and effort-intensive process for therapists and caregivers. Text-to-Image (T2I) models can generate diverse, realistic, and custom images from text prompts. In this paper, we explore the use of T2I models to generate image therapy material for aphasia. To this end, we used DALL-E 2, Stable Diffusion XL (SDXL), Stable Diffusion (SD) Turbo, and Flux Schnell to generate images of different resolutions. We quantify the resulting image quality using the Inception Score (IS), CLIPScore, and Text-to-Image Faithfulness Evaluation with Question-Answering (TIFA). Based on these metrics, experiments show that DALL-E 2 and SDXL typically render better images at high resolution (1024 \(\times \) 1024). Additionally, Flux Schnell and SD Turbo generally generate better images at lower resolution (512 \(\times \) 512). This quantitative analysis can serve as a foundation for further research to facilitate aphasia rehabilitation by using generative machine learning methods.