Well, for this specific picture, it was for a D&D campaign. This was the prompt I fed the computer:

Within the dimly lit chamber of the Tower of Sorrows, shadows twist and writhe, seemingly alive. The air is heavy with an otherworldly stillness, broken only by the echoing sounds of combat and the flickering of arcane energies

I have messed around with various prompting methods, but only two things truly matter. Subject & Theme. If you look at my previous prompt, I provide a Theme through various elements such as the lighting, the location, and the types of words used to describe the scene. From my test, I assume it takes these themes as a framework and then applies them to a subject element, in this case, “The Tower of Sorrow”

I have also found that trying differing prompts is far better than making the picture attuned to specific elements forcibly. One such example is forcing an image by giving it another image and asking it to manipulate it. Such as taking a picture of a baby and saying, “Make this picture, baby Yoda.” Although the results are intriguing and close, I feel that not giving the computer the appropriate theme or atmosphere creates complications for the AI to focus on, and from that, begins to blur unfounded elements into the picture. All in all, build a methodology you like and stick with the framing. That being said, it never hurts to throw random words in and see what kind of beauty comes out!