Write image prompts that give steady results

1. Why results feel random

A short request like “a nice photo of coffee” leaves the tool to guess everything. Each guess differs, so each image does too.

Fix this by deciding the same five things every time. Once those are fixed, your images look like a set.

2. The five parts

1. Subject: what is in the picture, in plain words.
2. Setting: where it is and what is around it.
3. Style: photo, flat illustration, sketch or another look.
4. Light and colour: soft daylight, warm tones, muted palette.
5. Shape: where it will be used, such as square or tall, and any space left for text.

3. Build and test yours

1. Fill in all five lines for one real image you need.
2. Generate it, then change only one line and generate again.
3. Keep the version you prefer and save the whole prompt in your notes.

Changing one line at a time shows you which part controls what. That is how you learn the tool quickly.

4. Use images safely

• Check the tool’s terms before using images commercially.
• Do not ask for a real person’s likeness without their permission.
• Check hands, text and small details, which often come out wrong.
• Say an image is AI-made where a platform or your customers expect it.

Common mistakes to avoid

• Writing one vague line and hoping
• Changing three things at once
• Asking for text inside the image and trusting the spelling
• Using a real person’s likeness without permission

Final checklist

• My prompt fills in all five parts
• I changed one line at a time
• I saved the prompt that worked
• I checked details and usage terms