How to Write AI Image Prompts: A Guide with Examples

The difference between a great image and a mediocre one usually isn't hidden in the model — it's hidden in the prompt. The same model produces ordinary results from a vague sentence, and striking ones from a clear, layered brief. In this guide, we'll walk you step by step through the method that gets you closest to the image in your head on the very first try with Oxava AI.

The anatomy of a great prompt

An effective prompt isn't a string of random words — it's a shoot brief. Think through five layers in order:

1. Subject — what are we creating? (a product, a person, a landscape, an object)
2. Setting / background — where? (studio, street, nature, abstract backdrop)
3. Light — how is it lit? (soft side light, golden hour, hard shadow)
4. Composition / camera — which angle and lens? (close-up, macro, wide angle)
5. Style and quality — which aesthetic? (photorealistic, cinematic, 4K, editorial)

When you combine these five layers into a single sentence, the model knows exactly what to imagine.

Bad prompt vs. good prompt

A vague request forces the model to fill in the gaps on its own:

❌ "A perfume image"

Describe the same idea in layers:

✅ "Matte black glass perfume bottle, on wet rocks, in morning mist, soft side light, macro lens, photorealistic 4K product photo"

Because the second prompt spells out subject, setting, light, lens, and quality one by one, the result is predictable and on-brand.

Be specific — show your adjectives

Abstract adjectives like "beautiful," "professional," and "high quality" tell the model very little. Replace them with visible detail:

• instead of "nice light" → "soft natural light from a window, gentle shadows"
• instead of "professional" → "editorial fashion shoot, clean background, 85mm lens"
• instead of "vivid colors" → "warm orange-teal palette, high contrast"

Choose the right model

Every model shines at something. On Oxava, pick the model based on the subject:

• For realistic portraits and people, use human-focused models (face, skin, expression detail).
• For products and e-commerce, use models with strong composition and text/label consistency.
• For fast variations and idea scouting, use lightweight, fast models.

Starting with the right model is more efficient than forcing the prompt. For tips specific to product shoots, check out our AI Product Photography article.

Use a reference image

If you have a real product, model, or logo on hand, upload it as a reference. The model preserves the form and identity and only rebuilds the scene — it's the safest way to try unlimited variations without breaking brand consistency. You can provide multiple references at once (e.g., one model + a few outfits).

Choose the aspect ratio to fit the job

Aspect ratio matters as much as resolution: square (1:1) for e-commerce cards, vertical (9:16) for stories, and horizontal (16:9) for covers and banners. The right ratio prevents loss from cropping later.

Strengthen your prompt with "Enhance"

If you're struggling to describe what you want, write a short idea and use the Enhance feature: it turns your brief sentence into a layered, model-friendly prompt. When you add a reference image, it also reads the image's contents and folds them into the brief.

Common mistakes

• Overloading: Ten different requests in a single frame scatter the model. Focus on one idea.
• Contradictory instructions: Opposites like "minimal but detailed, plain but flashy."
• Mistaking resolution for a prompt: "4K" boosts quality but doesn't build composition.
• Giving up after one try: The first output points the way; you close in with small tweaks.

The next step

Pick an idea, gather the five layers into a single sentence, and start generating. If you like the result, you can upscale the image up to 4K and get it print-ready. For inspiration, browse the examples on the Explore page, or head straight to the studio.