Lesson 1 of 5 · 12 min · Free
How to write AI image prompts that work
Learn how to write AI image prompts that work: a simple five-part structure (subject, style, composition, light, details) with example prompts to copy.
You typed a few words into an AI image generator and got something close but not quite right. The cat was there, but the light was flat and the background was busy. That's normal. The model filled every gap you left with its own guesses.
A good prompt leaves fewer gaps. This lesson gives you a simple five-part structure that works on any modern image model, plus examples you can copy and adapt.
Why short prompts give generic results
An image model has to decide many things to make one picture: the subject, the camera position, the time of day, the medium. When you write "a cat on a sofa," you've made two of those decisions. The model makes the rest, and it tends to choose the most average option.
You don't need a long prompt. You need a specific one. Specific nouns beat adjectives like "beautiful" or "amazing," which tell the model almost nothing.
The five parts of a prompt that works
Think of a prompt as a short brief for a photographer or illustrator. Cover these five things, roughly in this order.
1. Subject
Lead with what the picture is of, and be concrete. Not "a dog," but "a scruffy gray terrier puppy with one ear up." Not "a mug," but "a handmade speckled ceramic mug." If there's an action, include it: sitting, laughing, pouring.
2. Style or medium
Say what the image is made of. A photograph, a watercolor, an oil painting, a flat vector illustration, a vintage screen-printed poster, a 3D render. This one choice changes everything else, so don't leave it to chance.
3. Composition
Describe the framing the way a camera operator would. Close-up, wide shot, overhead, eye level. Where is the subject in the frame? What's in the background, and is it sharp or soft? If you need empty space for text, ask for it.
4. Light
Light sets the mood more than any other detail. "Soft morning light from the left," "overcast daylight," "warm lamp light at night," "harsh midday sun." Naming the direction and quality of light is the fastest way to make an image look less artificial.
5. Details and constraints
Finish with the small things that matter to you: colors, materials, props, and anything you want left out. Constraints are useful. "Clean and minimal, no other objects" or "natural skin texture, no retouching look" steer the model away from its habits.
Putting it together
Here's a vague prompt and the same idea written with all five parts.
A coffee mug on a table.
A product photo of a handmade speckled ceramic mug on a light oak table. Eye-level, the mug slightly off center, soft-focus background with a pine sprig. Soft morning light from the left. Clean and minimal, no other props, muted natural colors.
The second prompt just answers the questions the model would otherwise guess at. Two more to try:
A watercolor illustration of a red fox curled up asleep in fresh snow. Close-up, the fox filling the lower half of the frame, pale sky above. Cool blue dawn light. Loose brushstrokes, white paper showing through, limited palette of rust, blue and cream.
A vintage screen-printed poster with the headline "HOLIDAY NIGHT MARKET" in bold block letters, and a second line reading "DEC 12, 5 to 10 PM, MAIN STREET". Centered layout with string lights across the top. Deep green and cream only, grainy print texture.
Notice the poster prompt puts the exact words in quotation marks. When you want text in an image, spell it out and quote it. Then check the result letter by letter before you use it.
How to fix a result you don't like
Treat your first image as a draft. Look at it and ask which of the five parts went wrong, then change only that part. Changing everything at once makes it hard to learn what worked.
- It looks generic. Your subject or style is too vague. Add specific nouns and name the medium.
- It looks fake or plastic. Describe the light, and add a constraint like "natural texture, not overly polished."
- It added things you didn't want. Say what to leave out, plainly: "no people," "no other objects."
- It ignored an instruction. Repeat it more firmly, or move it earlier in the prompt.
That last tip comes from our test of GPT Image 2.5, Nano Banana Pro and Nano Banana 2. Every model broke one rule from the prompt in that test, and each mistake was fixable in one more run by restating the instruction it skipped.
Common mistakes to avoid
- Stacking adjectives ("beautiful, amazing, 8k") instead of describing what you actually want.
- Asking for several unrelated subjects in one image. Start with one clear subject.
- Forgetting the format. A vertical story and a wide banner need different compositions, so decide the shape first.
- Giving up after one try. Small revisions usually get you there.
Try it now
You can practice on Create, where you pick a model and write your prompt, or with the Text to Image tool, which adds style presets and aspect ratios if you'd rather not describe those in words. New accounts get 10 free credits, and credits never expire, so you can take your time.
In the next lesson, you'll learn how to choose between models so you're spending credits where they make a difference.
Practice
Pick one everyday object near you and write a three-word prompt for it, such as "a plant pot." Generate it once. Then rewrite the prompt using all five parts: a specific subject, a medium, a camera framing, a named light source and one constraint about what to leave out. Generate again, compare the two, and note which single part made the biggest difference.