The Top Prompt Mistakes (and Their Fixes)
Overstuffing, contradictions, vague adjectives and ignored aspect ratios — the four failure modes behind 90% of disappointing generations.
You now own the full toolkit: anatomy, style words, camera language, iteration. This final lesson is the crash-report review: the four mistakes that cause almost every disappointing generation, and the mechanical fix for each. Learn to spot them in other people's prompts and you will stop writing them yourself.
Mistake 1: overstuffing (keyword soup)
Piling on every buzzword you have ever seen — masterpiece, trending on artstation, 8K, hyperdetailed, award-winning — does not stack quality. Attention is a fixed budget: every filler word taxes the words that matter, and past roughly 60-70 words models start silently dropping instructions. If a word does not change pixels, it is dead weight.
Before: the overstuffed prompt
masterpiece, best quality, ultra realistic, 8K, hyperdetailed, award winning photo, trending on artstation, epic, stunning, beautiful girl, perfect face, perfect lighting, cinematic, dramatic, moody, vibrant colors, dark tones, intricate details, sharp focus, bokeh, wide angle, close-up portrait
Count the actual scene content: one vague subject. Everything else is quality-word confetti — plus two hidden contradictions we will meet next (vibrant/dark, wide angle/close-up).
Mistake 2: contradictions. 'Golden hour sunlight under a starry midnight sky.' 'Minimalist scene, packed with intricate details.' '16mm wide vista, tight close-up.' The model cannot satisfy both, so it averages them — and the average of two moods is no mood. Contradictions usually sneak in when you bolt new ideas onto an old prompt without rereading it. Reread before every run.
Mistake 3: vague adjectives. 'Beautiful', 'epic', 'stunning', 'high quality' are opinions, not descriptions — they name the reaction you want, not the pixels that cause it. The fix is the swap you learned in scene description: replace every judgment word with the visible evidence for it.
| Mistake | Fix |
|---|---|
| masterpiece, 8K, best quality (x12) | cut them; add one concrete style anchor instead |
| golden hour + starry midnight sky | pick one time of day and commit |
| a beautiful epic landscape | jagged dolomite peaks above a wildflower meadow, storm light |
| wide vista + extreme close-up | choose the shot: one framing per image |
| square logo prompt left at 16:9 | set aspect ratio to 1:1 before generating |
Mistake 4: ignoring aspect ratio
Aspect ratio is not a crop you apply later — the model composes into the frame you give it. A full-body character in 16:9 becomes a tiny figure lost in empty landscape; a panoramic skyline in 9:16 gets amputated. Match the frame to the subject before you generate: 9:16 for standing figures and phone stories, 1:1 for logos and products, 16:9 for landscapes and thumbnails.
After: everything fixed
Close-up portrait of a freckled violinist backstage, eyes closed mid-note, single warm spotlight from the left carving her profile out of deep shadow, dust motes in the beam, 85mm, f/1.8, Ilford HP5 black and white, 4:5 portrait
32 words, zero filler, zero contradictions: one subject, one light, one mood, one framing, and an aspect ratio that matches the shot. This is the overstuffed prompt's redemption arc.
That closes the Prompt Engineering path — you can now out-prompt 95% of users on any model. Next stop: applying it all with real tools and real pixels in the Image Generation path, or browse the prompt library to reverse-engineer prompts you admire.
Put the checklist to work on a real brief: pick a template and make it yours with a clean, contradiction-free prompt. Browse Templates