Short answer: describe the subject, the setting, the style, the lighting and the framing, in that order, in plain sentences. Be specific about what matters, say where any text goes, and refine in small steps instead of rewriting the whole prompt.
The five-part prompt formula
A reliable prompt answers five questions:
- Subject: who or what is in the image, and what they are doing.
- Setting: where it is, and when.
- Style: photo, illustration, 3D render, watercolor, flat vector.
- Light and mood: soft morning light, neon at night, overcast and calm.
- Framing: close-up, wide shot, top-down, and the aspect ratio you need.
Strong: "A small neighborhood coffee shop at 7am, a barista pouring latte art behind a wooden counter, warm window light, shallow depth of field, documentary photo style, horizontal 16:9."
Nine tips that work
1. Write sentences, not keyword lists
Modern models like ChatGPT Images and Nano Banana Pro understand natural language well. "A red bicycle leaning against a blue door" works better than "bicycle, red, door, blue, 8k, masterpiece".
2. Put the most important thing first
Models pay more attention to what comes early. Lead with the subject.
3. Be specific about the details you care about
If the color of the jacket matters, say so. Anything you leave out, the model will choose for you.
4. Name a medium, not a living artist
"Risograph print" or "1970s travel poster" gets you a style without copying a specific person's work, and some tools block artist names anyway.
5. Put text in quotes and say where it goes
For posters and social graphics, write: the headline "Summer Sale" in bold sans-serif at the top. Ideogram is the specialist for text-heavy images, and ChatGPT Images is also strong at it.
6. State the aspect ratio
Say 1:1 for social posts, 9:16 for stories, 16:9 for thumbnails and headers. Cropping later loses detail.
7. Refine in small steps
When the image is close, ask for one change: "same image, but make the sky overcast". Conversational editors like Nano Banana Pro are especially good at this.
8. Use a reference image when words fail
Most tools accept an image upload. A reference for composition, color or product shape saves many rounds of prompting.
9. Describe what you want, not what you do not want
"An empty street" works better than "a street with no people", because mentioning people can bring them in.
Prompt examples by use case
- Product shot: "A matte black water bottle on a light gray stone surface, soft studio light from the left, clean minimal background, product photography, square 1:1."
- Blog header: "Flat vector illustration of a person organizing floating folders in the cloud, limited palette of navy, lime green and white, lots of empty space on the right, 16:9."
- Logo idea: "Minimal geometric logo mark for a bakery called Crumb, a simple wheat stalk forming the letter C, single color, vector style, white background." Recraft is best when you need editable vectors.
- Poster: "Retro 1970s travel poster of a mountain lake at sunset, grainy print texture, the title "Lake Days" in large rounded letters at the top, portrait 2:3."
Fixing common problems
- Hands or faces look wrong: regenerate, or ask for a different pose or a wider shot.
- Text is misspelled: shorten it, put it in quotes, or use Ideogram.
- The style keeps drifting: repeat the style words at the end of each edit, or reuse your first image as a reference.
- Everything looks generic: add specific details about place, time and materials.
Which tool should you use?
ChatGPT Images is the best all-rounder and the easiest way to get what you asked for. Midjourney leads for artistic style, Adobe Firefly is the cautious commercial choice, and FLUX.2 suits developers who want photorealism. Compare all seven in our ranking of the best AI image generators.
See every option ranked side by side, with prices, free plans and honest downsides.
See the top 7 AI image generators →Frequently asked questions
What is a good AI image prompt?
A good prompt describes the subject, setting, style, lighting and framing in plain sentences, with specific details for anything that matters to you, plus the aspect ratio you need.
Do long prompts work better?
Specific prompts work better, not necessarily long ones. Two or three clear sentences usually beat a long list of keywords.
Why does AI misspell text in images?
Text is still one of the hardest parts of image generation. Keep text short, put it in quotes, say where it goes, and use a tool that is strong at typography like Ideogram or ChatGPT Images.
Can I use AI images commercially?
It depends on the tool and plan. Read each tool's terms, and choose a commercially cautious option like Adobe Firefly if usage rights matter to your business.
Prices and plans were checked in November 2026 and change often. How we research and rank tools.