The easiest way to get caught when writing custom prompts is when everything outside the main subject is empty. A lot of people write super detailed character descriptions and actions, but then just slap some template sentence at the end for background, lighting, or camera angle—you can tell it’s copy-pasted from a mile away. What I do now is flip it around: first, I think about what scene this image is supposed to land in—is it an e-commerce white background, a magazine spread, or a game concept art? Nail down the medium and purpose, then fill in the subject. That way, every sentence serves that specific need, so it doesn’t feel like an assembly line. Piling on adjectives is useless—you need every word to have a reason it absolutely has to be there.
Yeah, flipping it around and setting the scene first works like a charm. I used to just cram the prompt with the subject and leave the background totally empty.
Once the medium is locked in, even the camera angle is easy to pick—white background just means flat lighting.
Stacking a bunch of adjectives is useless. I cut half of them out and it actually got more accurate.
“Every word needs a reason” — that’s spot on. Now whenever I finish writing, I ask myself what would happen if I cut that word.
Magazine spreads and game concept art are two completely different beasts, you can’t just reuse the same templates.
Tried putting the use case at the start of the prompt, and the image direction immediately converged.
But sometimes having too rigid a scene kills the surprise, so I’ll leave one or two prompts loose.
E-commerce white backgrounds eat this shit up, any random copied background looks fake as hell at a glance.
Oh yeah, I’m all over this one.