Consider reducing image dimension to 512x512 or 512x768. Stable Diffusion is trained on images of that size & it may have difficulty filling the additional space, thereby adding in duplicates or multiples of heads, people, objects, etc.
In addition, try using negative prompts like: duplicate, copy, multi, multiple, two faces, two, disfigured, etc. You can also specify "single" or "one" in your actual prompt.
Keep re-trying generations even if you're not successful at first - it sometimes takes multiple tries.
Note, you're still able to use the 4x upscale feature to generate higher resolution images once you get something you like created.