How to write AI image prompts that actually work
Effective AI image prompts read like a short creative brief: the subject and what it is doing, the setting, composition and camera angle, lighting, style and colour, and the format or aspect ratio. Add reference images where the tool supports them, state what to avoid, and iterate by changing one element at a time. Save prompts that work in a shared library so your team gets consistent, on-brand results across tools such as Midjourney, Adobe Firefly, Google Imagen and ChatGPT.
Why do so many AI images look generic?
Short prompts such as “a coffee cup on a table” leave the model to guess almost everything, so it falls back on average, stock-like results. Professional photographers and art directors work from briefs that specify composition, light and mood. Writing prompts the same way produces more distinctive, usable images. Our comparison of the best AI image generators covers which tools suit which jobs.
What should a prompt include?
| Part | What to write | Example |
|---|
| Subject | What is in the image and what it is doing | A barista pouring latte art |
| Setting | Where and when | A small Dubai café at sunrise |
| Composition | Framing and angle | Close-up, eye level, subject on the right third |
| Lighting | Source and quality | Soft window light, gentle shadows |
| Style and colour | Look and palette | Editorial photography, warm neutral tones |
| Format | Aspect ratio and use | 4:5 for Instagram, space for a headline |
How do you describe style without copying artists?
- Describe the medium: editorial photography, watercolour, flat vector illustration, 3D render
- Describe the mood: calm, energetic, luxurious, playful
- Describe colour: muted pastels, high contrast, monochrome with one accent
- Describe lens and light: shallow depth of field, golden hour, studio softbox
- Avoid naming living artists; describe the qualities you want instead
How do reference images help?
Many tools accept reference images for style, composition or a specific subject. Upload your product photo to keep it accurate, a brand photo to match lighting and colour, or a layout sketch to guide composition. References are the most reliable way to keep a series consistent, especially for products and recurring characters.
How do you tell the model what to avoid?
Some tools support negative prompts or explicit exclusions; others respond better to positive descriptions. Where supported, list what you do not want, such as text, watermarks, extra people or clutter. Otherwise, describe a clean scene positively: “a minimal, uncluttered table with a single cup”.
How do you refine results?
- Generate several options: Start with four to eight variations.
- Pick the closest: Choose the result that best matches the brief.
- Change one thing: Adjust lighting, angle or style, not everything at once.
- Use editing tools: Inpaint or use Generative Fill to fix small areas.
- Finish in your design tool: Crop, colour-correct and add text in Photoshop, Figma or similar.
What does this look like in practice?
A café brand needs a set of Instagram images for a new seasonal drink. The designer writes a base prompt: an iced saffron latte in a clear glass on a light stone counter, shot from a slight high angle with the glass on the right third, soft late-afternoon window light with long shadows, minimal editorial style with warm cream and gold tones, 4:5 format with empty space on the left for text. She uploads a photo of the real glass as a reference. After three rounds, one adjusting the angle and one softening the shadows, she has a consistent set of four images and adds headlines and the logo in Figma.
Why keep a prompt library?
Save every prompt that produces an approved image, with the tool, settings, references and a thumbnail. Group prompts by use, such as product, lifestyle and backgrounds. A shared library helps the whole team produce consistent, on-brand images quickly and makes onboarding new designers easier. Our guide to AI-ready design systems explains how written rules help AI tools stay on-brand.
What mistakes should you avoid?
Most poor results trace back to a handful of habits. Avoiding them saves more time than any clever prompt trick, and makes results far more predictable for the whole team.
- Stacking dozens of keywords instead of describing a scene
- Changing several things at once between rounds
- Relying on AI for logos and exact product details
- Ignoring aspect ratio until the end
- Skipping a final design review
- Forgetting to save the prompts that worked, so the team has to rediscover them next time
How UIUX Studio helps
We help teams build prompt libraries and AI workflows that fit their brand. Talk to our product design team. For content creation at volume, Social Mafia by UIUX Studio produces branded images and social content.
Frequently asked questions
How long should an AI image prompt be?
Usually one to three sentences covering subject, setting, composition, light, style and format.
Do prompts work the same in every tool?
The structure works broadly; each tool interprets details differently.
Should I name artists in prompts?
Describe the qualities you want instead of naming living artists.
How do I keep a product accurate?
Use a reference image of the real product and check every result.
Can AI add text to images?
Some tools can; for brand work, add text in a design tool.
Related articles