Breast thermography is gaining renewed interest as a non-invasive, radiation-free, and privacy-conscious method for assessing breast health. This resurgence is largely driven by advancements in artificial intelligence (AI). However, the development and validation of AI algorithms for breast thermal imaging are hindered by the lack of large, high-quality datasets. Generative AI algorithms have demonstrated significant potential in producing high-fidelity images, yet research on synthesizing breast thermal images remains limited. In this study, we evaluate and compare the performance of three deep learning architectures: Deep Convolutional Variational Autoencoder (DCVAE), Deep Convolutional Generative Adversarial Network (DCGAN), and Denoising Diffusion Probabilistic Models (DDPM) in generating high-fidelity synthetic breast thermal images. Our results demonstrate that DDPM achieves the highest fidelity in image generation, with a Fréchet Inception Distance score of 18.1, surpassing DCGAN and DCVAE variants, which achieve scores of 129.7 and 150.8, respectively. These results underscore the potential of diffusion networks in advancing synthetic image generation for breast thermography and addressing the challenges posed by limited real-world datasets.