How do you turn a drawing prompt into a Photoshop practice exercise?

Always with a visual reference, often with multiple visual elements — Generative Fill is more predictable than Generate Image IME.   I find visual references generally work better when placed on a white background on the original canvas, rather than as an uploaded reference image.  The Ai models tend to change reference images too much.   I like to sneak up on a complex image as it helps you keep control.  So this render from the current Something For The Weekend thread was teased towards the abbey walls on the grassy mound.  I added my avatar image on whitespace below and was able to complete the image with a single prompt.  It doesn’t matter how you lay out the visual elements. You just need to be able to describe them and their position in the Prompt. i.e. ‘The cartoon man at the bottom of the original image...’