How to Write Better AI Image Prompts
Six building blocks that turn a vague idea into a picture the model actually understands.
Most disappointing AI images come from the same root cause: the prompt left too much to chance. The model isn't reading your mind — it's filling every gap you leave with its own averages. The fix isn't longer prompts or magic words. It's being deliberate about a handful of things the model genuinely pays attention to.
Here is a reliable structure you can reuse for almost any image, whether you're on Midjourney, FLUX, Nano Banana, Seedream, or DALL·E.
1. Subject: say exactly what (and who) is in the frame
Start with the single most important thing the image is about, described concretely. "A woman" is weak; "a middle-aged woman with short silver hair wearing a wool coat" gives the model something to hold onto. Include count ("three lanterns," not "some lanterns"), and any details that matter to you — if you don't specify them, you're rolling dice.
2. Style: name the look you want
Style is where people are vaguest, and it's the biggest lever. Decide whether you want a photograph, an oil painting, a 3D render, a flat vector illustration, anime, a pencil sketch, and say so. You can go further with era and artist-movement references ("1970s film photography," "Art Nouveau poster," "cyberpunk concept art"). If you want realism, the word photograph or photorealistic does more work than any other single term.
3. Composition: describe the shot, not just the scene
Think like a photographer or director. Where is the camera? How much of the subject is in frame? Useful terms include close-up, wide shot, overhead / top-down, low angle, portrait orientation, rule of thirds, and shallow depth of field. "A cabin in the woods" and "a wide, low-angle shot of a cabin in the woods, tall pines framing the edges" produce very different results.
4. Lighting: the fastest way to change the mood
Lighting quietly decides whether an image feels warm, dramatic, clinical, or dreamy. A few dependable choices: golden hour for warm and flattering, soft diffused light for gentle and even, hard directional light for high contrast and drama, backlighting for silhouettes and glow, neon for moody urban scenes. Even adding just "soft window light" to a portrait prompt noticeably lifts the result.
5. Mood and color: set the emotional tone
Tell the model how the image should feel and roughly what palette to lean on: "moody and melancholic, muted blues and greys," or "cheerful and bright, warm pastel palette." This ties the other elements together and stops the model from defaulting to a generic, over-saturated look.
6. Details and constraints: last, and optional
Finish with any extra specifics — textures, background elements, weather, small props — and, on models that support them, technical parameters like aspect ratio. Keep these at the end so they refine rather than dominate the prompt.
Putting it together
Stacking those blocks might give you: "A photorealistic close-up portrait of a middle-aged woman with short silver hair, wearing a wool coat, soft window light from the left, shallow depth of field, calm and contemplative mood, muted earthy tones." Every clause there is a decision you made instead of leaving to the model.
Iterate, don't restart
When a result is close but not right, change one thing at a time. Swap the lighting, tighten the composition, or adjust the mood — then compare. Rewriting the whole prompt from scratch throws away information about what was already working. Small, controlled edits are how you steer toward the picture in your head.
Want a head start? Our free Prompt Generator walks you through these same building blocks and formats the result for the specific tool you're using.