As usual, Midjourney decides to do the opposite of what I tell it to do (run away, not towards it!)
I haven't used Midjourney, but one tip that seems to work for me in Stable Diffusion when I want a body part included is to mention it (e.g. "black hair, sneakers" will tend to get a person in the frame from head to feet). Like, maybe try putting in text describing her face, say.
Also, it looks like Midjourney supports inpainting. It's more time-consuming, doesn't let you easily go back and twiddle the image as a whole, but you can go and erase the objectionable part of an image and regenerate it with a different prompt that doesn't affect the rest of the image.