Increase the resolution. Even when that shouldn’t help, it does, if that makes sense.
In the negative prompt, describe types of images that aren’t photos. Anime, render, cartoon, etc.
In the positive prompt, describe a detail that you would only find in a highly detailed image. Some people just write “highly detailed,” but that does jack shit because no one’s out there tagging photos as “highly detailed.” Or they’ll say “8k resolution,” which works, but it’s too limiting and it’ll throw off the style too much. In this case, I included “sweaty skin.” A photo that captures the texture of skin is going to be highly detailed, and those photos will become the AI’s source material. This has the side effect of making the composition rather tightly framed, but that worked here.
A few things that I’ve found help: