A Reproducible Extraction of Training Images from Diffusion Models