Finally got decent results from an AI image generator. Here's what actually matters in the prompt.
After weeks of struggling with vague prompts and disappointment, I finally figured out what actually works. I kept reading guides about adding art styles and "4k, cinematic lighting, trending on Artstation" (which is cringy anyway), but that wasn't the real issue.
What actually matters: BE SPECIFIC about composition and lighting. Instead of "fantasy warrior," try "warrior shot from below at sunset, golden light hitting their face from the left, dark background." Instead of just naming an art style, describe actual visual qualities: "soft lighting," "high contrast," "muted colors."
Also, negative prompts are your friend. Excluding things you don't want ("no blurry hands, no extra fingers") genuinely improves results more than adding more stuff to your positive prompt.
I'm not claiming I'm an expert or anything, but if you're frustrated with results, maybe try being aggressively specific about the visual qualities you actually want instead of just describing the subject. Would be curious if this approach works for other generators too- I've mainly used Fiddlart.