Writing Image Prompts That Work

You type "a podcast cover about money" and get a pile of gold coins floating in space. You try again with "a better podcast cover about money" and get a different pile of gold coins. The tool is not ignoring you. You gave it four words, and it filled every other decision with the most common guess: what the scene looks like, what style, what colors, how it is framed, how it feels.

Every image has dozens of choices baked into it. If you do not make them, the model makes them for you, and its default is usually the most generic version of your idea. A strong image prompt makes the choices that matter to you, in plain words, so the model has less to guess.

This lesson gives you a reusable structure, shows it working on a real example, and teaches the habit that separates people who get good images from people who keep rerolling: change one thing at a time.

The building blocks

You will not need every part every time. But when an image disappoints you, one of these is almost always the part you left to chance, so the table doubles as a checklist.

A few notes on each

Subject comes first and should be concrete. "Success" is an idea; "a woman crossing a finish line with her arms raised" is a picture. Describe what the camera would actually see.

Style or medium often changes the result more than anything else. The same subject as a "35mm film photograph" and as a "children's book illustration" are two different images. If you only add one thing to a weak prompt, add this.

Composition matters when the image has a job. If you plan to put a title over it, ask for "plenty of empty space at the top." If it will be a small thumbnail, ask for "a single bold subject, simple background."

What to avoid is useful, but keep it short and specific. A long list of things you do not want can pull attention toward them. "No text in the image" is one of the most useful avoid-lines you will ever write, since lettering is a weak spot for these tools.

Aspect ratio depends on the tool. If there is a setting for image shape, use it rather than describing the shape in words. If there is not, choose a composition that will survive cropping, and crop afterward.

A template you can reuse

[SUBJECT], [SETTING].
Style: [STYLE OR MEDIUM].
Composition: [FRAMING AND CAMERA ANGLE].
Lighting: [LIGHTING].
Mood: [MOOD].
Colors: [COLOR PALETTE].
Avoid: [WHAT SHOULD NOT APPEAR].

Short labeled lines like these are easier for you to edit later, and they make it obvious which part you changed between runs. Plain comma-separated sentences also work. What matters is that each decision is stated somewhere.

Worked example: a podcast cover

Marcus hosts a weekly podcast called "Paycheck to Plan," about personal budgeting for people who have never had a budget. He wants cover art that feels encouraging, not like a bank ad. He will add the show title himself in a design tool.

Weak prompt:

podcast cover about money

The result: coins, dollar signs, maybe a piggy bank, in a glossy style. Nothing about it says "calm and doable," and it looks like every other finance show.

Stronger prompt:

A young man at a small kitchen table writing in a notebook, a mug of
coffee beside him, in a modest apartment in the evening.
Style: warm flat illustration with simple shapes and subtle texture.
Composition: square image, man in the lower half, lots of empty space
in the top third for a title.
Lighting: warm lamp light.
Mood: calm, capable, quietly hopeful.
Colors: deep teal, mustard yellow, and cream.
Avoid: no text, no coins, no dollar signs.

Now the image tells a story about a person getting organized, which is what the show is actually about. The empty space at the top gives Marcus room for his title. The avoid-line removes the clichés. The palette gives him colors he can reuse on episode graphics.

Is it perfect on the first run? Probably not. Maybe the man looks too stressed, or the illustration is busier than he wanted. That is where iteration comes in.

Iterate one element at a time

When an image is close but not right, the tempting move is to rewrite the whole prompt. Resist it. If you change five things and the next image is better, you do not know which change helped. If it is worse, you do not know which change hurt.

Instead:

1. Name the one thing that is wrong. "The mood is too serious," not "I don't like it."
2. Find the part of the prompt that controls it. Mood, lighting, colors, style.
3. Change only that part and run it again.
4. Compare the two results side by side. Keep the version that is better, then move to the next problem.

For Marcus, the sequence might look like this: first run, the man looks worried, so he changes the mood line to "relaxed, a slight smile." Second run, better, but the background is cluttered, so he adds "simple background with only a window and a plant." Third run, the colors are right but the style is too detailed, so he changes the style line to "minimal flat illustration, very few details." Three runs, three specific changes, and he knows exactly why the final one works.

Remember that each run builds a new image, so some variation between runs is normal even when you change nothing. If a change seems to make no difference, run it once more before deciding it does not matter.

Try this

Give yourself 10 to 15 minutes in the Playground.

1. Pick a real image you need: a post graphic, a slide background, a worksheet illustration, a cover.
2. Write the weak version first, in five words or fewer, and run it. This is your baseline.
3. Fill in the template above for the same image and run it.
4. Look at the result and name the single biggest problem in one sentence.
5. Change only the line that controls that problem. Run it again.
6. Repeat step 4 and 5 once or twice more. Stop when you have an image you would actually use, or when you have made three changes.
7. Favorite the best result in your Library, and copy your final prompt somewhere you keep useful prompts, with [BRACKETS] replacing the details that will change next time.

Every run counts toward your monthly image allowance, so a planned sequence of single changes gets you further than rerolling at random.

What to remember

• A short prompt leaves most decisions to the model, and its default is the most generic version of your idea.
• Use the building blocks as a checklist: subject, setting, style, composition, lighting, mood, colors, what to avoid, and shape if the tool supports it.
• Style or medium often changes an image more than anything else. If you add one thing, add that.
• Plan for text you will add later: ask for empty space and "no text in the image."
• Iterate one element at a time so you know what each change did.