AI images become useful when you can explain why one version fits the job better than another. This walkthrough starts with a brief, creates a small set and records each decision. It works with any image tool that accepts text instructions.
Your finish line: one selected image, one rejected alternative and an edit log another person can follow.
Start with the placement, not the style
Write where the image will appear before describing what it should look like. A homepage hero needs room for text. A square social post needs a readable subject at thumbnail size. A product concept may need a neutral background so details are easy to compare.
| Brief field | Example | Why it matters |
|---|---|---|
| Placement | Website hero, 16:9 | Sets the frame and empty space |
| Subject | Ceramic desk lamp | Keeps the object stable |
| Audience | People furnishing a home office | Guides the mood without inventing a persona |
| Must keep | Warm orange shade, no text | Creates objective checks |
| Can vary | Camera angle and surface | Gives exploration a boundary |
Build the first prompt
Use a plain sequence: subject → environment → composition → light → color → constraints.
A warm-orange ceramic desk lamp on a walnut desk in a quiet home office.
Wide website hero composition with the lamp on the right and clear negative space on the left.
Soft morning window light, restrained charcoal and warm-orange palette, realistic materials.
No words, logos, extra lamps or people. 16:9.Generate a small batch. Do not change the prompt between images yet; first learn how much variation the tool produces on its own.
Choose with a scorecard
| Check | Keep | Revise | Reject |
|---|---|---|---|
| Brief | All required elements are present | One fixable miss | Wrong subject or placement |
| Composition | Text area stays clear | Crop may solve it | No usable empty space |
| Detail | Looks plausible at intended size | Minor artifact | Broken shape or misleading detail |
| Brand fit | Palette and mood fit | Color adjustment needed | Conflicts with the brand |
The most dramatic image is not automatically the best asset. Select the one that survives the checks at its real display size.
Revise one variable at a time
Keep the subject and composition fixed while changing one item: lighting, camera height, surface or color. Save the prompt beside each output. If you change five things at once, you cannot tell which instruction helped.
Should I name an artist in the prompt?
Describe observable qualities—medium, palette, lighting, composition and mood—instead. It gives you more control and avoids making a living creator’s name the entire direction.
Can I use the first image commercially?
Do not assume that. Check the current tool terms, your rights to uploaded references, brand rules and any people, products or marks visible in the result before publishing.
Why does an edit affect more than the selected area?
Generative edits can reinterpret nearby pixels. OpenAI’s current help page explicitly notes that selections may not be precise. Compare the whole image after every edit, not only the highlighted region.