This is expected. Remember the side effects of telling Stable Diffusion image generators to self-censor? Most of the images started being of the same few models.