Image-generation prompts
9 decisions a prompt of this kind should make on purpose, and 7 things worth leaving out. One of them is something only you can supply — we leave it blank rather than let a model invent it.
image, picture, photo, art, draw, illustration, render, midjourney, stable diffusion, dalle, dall-e, generate an image, wallpaper, logo, poster, concept art, portrait, photorealistic, anime, aspect ratio, --ar
- _Subject + specificityOnly you know this. Describe the subject itself — appearance, age or material, what it is doing, what expression it wears. Two or three concrete details.
- +Style / mediumInfer from intent: photorealistic for real scenes, clean flat vector for logos, cinematic painting for characters.
- +Composition / shot / framingMedium shot, eye-level, subject slightly off-center; close-up for portraits, wide for environments.
- +Lighting + moodSoft golden-hour for people/scenes; clean even studio lighting for product/logo work.
- +Setting / backgroundAdd a simple fitting environment; plain seamless background for logos/products.
- +Color paletteHarmonious, slightly muted palette unless the subject implies specific colors; echo brand colors for brand work.
- +Aspect ratioMatch to use: 16:9 scenes/wallpaper, 9:16 phone, 1:1 avatars/logos; default 16:9.
- +Detail / render cues'Highly detailed, sharp focus' — a couple of quality cues, not a keyword pile-up.
- +Negatives / exclusions'No text, no watermark, no extra limbs, no distortion' as a baseline for figurative work.
Left out for this type: role prompting (persona leaks into the scene as a token); prose format instructions ('respond clearly'); politeness/meta words parsed as content tokens; narrative explanation of why you want it; vague superlatives alone ('make it amazing'); keyword-stuffing dozens of style/quality tags; writing-task tone words ('friendly, conversational')