Hi! Yes it seems indeed to be the case that "lesser known" words or people or objects, don't seem to get as good image generation results as "more popular" ones. This is a direct effect of the training data the model was trained on, it simply doesn't recognise the prompt as it has fewer similar patterns stored in its training data and model.

However, you can install your own version of Stable Diffusion and train your own models, for example on Sikhs. There is a lot of information out there about "Stable Diffusion", and basically you only need a computer with a decent / modern Nvidia GPU (which could be at your home or somewhere in the cloud).