# How to Write AI Image Generation Prompts for Product Photos (2026)

> How to write AI image generation prompts for usable product photos: a four-part formula, the words that move pixels, worked examples and a fix list for drift.

Published: 2026-10-13
Updated: 2026-10-13
Author: Olga Stogova
Tags: ai-prompts, ai-product-photography, tutorial, prompt-writing, ecommerce
Canonical: https://kive.ai/learn/how-to-write-ai-image-generation-prompts

Learning how to write AI image generation prompts comes down to one habit: describe a photograph, not a feeling. A usable prompt names the subject, the objects around it, the direction and quality of the light, and the crop, in that order and in one sentence. Everything else, from mood words to lists of what to avoid, is noise the model resolves by guessing. This guide gives the formula, the words that work, and what to change when the result drifts.

## Key takeaways

- Write prompts in four parts in one sentence: subject, context, style, then one lighting or camera detail, subject first.
- Name objects and materials, matte black marble and gold confetti, not adjectives such as luxurious or premium.
- Say where the light comes from; light direction is the single word choice that most changes a product photo.
- Describe empty space and crop in words; it is how you reserve room for copy and get usable 1:1, 4:5 and 16:9 versions.
- Change one part per attempt; a prompt rewritten end to end teaches you nothing about which word did the damage.

---

## A prompt is a description of a photograph

An AI image prompt works when a photographer could set it up from the text alone. Subject, what is around it, what the light is doing, where the camera sits. That is the whole discipline, and it is why the prompts that look impressive, the ones stuffed with words like cinematic, ultra-detailed and 8k, produce less consistent results than a plain sentence a stylist would understand.

The second thing to know is where the words go. Image models weight the start of a prompt more than the end, so the subject belongs first and the technical detail last. A prompt that opens with a lighting mood and mentions the product in the final clause produces a lighting mood with a product somewhere in it.

The third thing is what a prompt cannot do. It cannot hold a product. A label, a cap, the exact green of a sneaker's heel tab are not describable in words to the precision a buyer needs, which is why product work splits into two jobs: a photo that anchors the product, and a prompt that describes the scene. Kive's own [prompt-writing guide](https://kive.ai/docs/creating-with-ai/writing-prompts) uses the same split, with the product attached as a saved model and the prompt reserved for context, style and light.

---

## The four-part formula, step by step

The formula is subject, context, style, technical detail, compressed to one sentence. The steps below build a prompt for a dropper bottle; swap the nouns and it works for any product.

### Build the prompt

**Name the subject first, as an object**

Start with the thing in the frame and its placement: an amber glass dropper bottle, upright, slightly left of center. If the product is anchored by a photo, this is where you attach it; if it is not, keep the description to shape, material and color and expect the label to drift.

**Add the context as a short list of objects**

Say what is physically around the product: on stacked gray stones, a warm circular backdrop behind. Objects and materials beat adjectives every time. Luxurious tells the model nothing; matte black marble and a satin ribbon tell it exactly which pixels to draw.

**Set the style with one reference, not five**

One style phrase is enough: editorial still life, clean e-commerce, moody campaign. Two styles fight. If you have a reference image whose look you want, most tools now accept it directly, and a reference does more than any style word can.

**Finish with light and crop**

Say where the light comes from and how hard it is, then the crop: warm directional light from the right with soft reflections, 4:5 portrait, product filling a third of the frame. Light direction is the detail that most changes a product photo, and it is the one most prompts leave out.

**Generate, look, change one thing**

Run the sentence. If the result drifts, change one part and run again. Swapping the whole prompt after a bad result is the most common mistake in prompt writing: you get a different image and no idea which word fixed or broke it.

Assembled, the prompt reads: an amber glass dropper bottle upright and slightly left of center, on stacked gray stones with a warm circular backdrop, editorial still life, warm directional light from the right with soft reflections, 4:5 portrait. Thirty-two words, four parts, nothing to argue with.

---

## Words that work and words that do nothing

Most prompt vocabulary falls into two piles: words that map to something a camera or a stylist controls, and words that describe how you want to feel about the result. The first pile moves pixels; the second is decoration.

| Part | Does nothing | Moves pixels |
| --- | --- | --- |
| Subject | A beautiful premium bottle | An amber glass dropper bottle, upright, slightly left of center |
| Context | Luxurious elegant setting | On a matte black marble pedestal, gold confetti out of focus behind |
| Style | Cinematic, ultra-detailed, 8k, masterpiece | Editorial still life |
| Light | Perfect lighting, dramatic | One warm rim light from the left, deep shadow on the right |
| Crop | High quality, professional | 4:5 portrait, product filling a third of the frame, empty space right |
| Exclusions | No text, no blur, no extra objects | Clean empty dark surface on the right, sharp edges on the product |

**Say what you want, not what to avoid**

Current image models respond better to positive phrasing than to exclusion lists. Instead of no blurry text, ask for sharp lettering or, better, for a clean empty surface where text would have been. Kive's documentation gives the same advice: describe the desired outcome rather than the thing to exclude.

_Context written as a list of objects: driftwood, stone, botanicals, nothing else_

---

## Worked prompts for four product categories

The formula holds across categories; the nouns change. These four are written for a sale season because dark scenes are the hardest to get right, and a prompt that survives black backgrounds survives anything. Each one leaves space for copy.

| Category | Prompt |
| --- | --- |
| Jewelry, fragrance, skincare | The product upright and off-center left on a matte black marble pedestal, soft gold confetti out of focus behind, editorial still life, warm rim light from the left, 4:5 portrait with empty dark space on the right. |
| Tech, gaming, streetwear | The product centered on a wet reflective black floor, red and blue neon tubes framing the background, light haze, high-contrast campaign look, hard light from above, 1:1 square. |
| Apparel, home goods | The product on a geometric stone podium in a dark charcoal studio, nothing else in frame, clean e-commerce look, one softbox from the left with a long soft shadow, 4:5 portrait. |
| Food, gift sets, cosmetics | The product on a dark wooden table, matte black gift boxes with satin ribbon blurred behind, warm fairy-light bokeh, festive still life, overhead beam on the product, 16:9 landscape with the product on the left third. |

Read any of them against the formula and the parts line up: object, objects around it, one style phrase, light direction, crop. The seasonal versions of these, with the tricks that make them hold up at phone size on a sale weekend, are in [the five prompt tricks for Black Friday](/learn/black-friday-prompt-tricks-ai-product-photos), and the full workflow from source photo to exported set is in [Black Friday product photos with AI](/learn/black-friday-product-photos-ai).

---

## When the result drifts, change the part that owns it

Every common failure maps to one part of the prompt, which is the practical reason to write in parts. Fix the part, not the sentence.

| What went wrong | Part that owns it | Change |
| --- | --- | --- |
| The product changed shape, label or color | Subject | Anchor it with a photo; if you cannot, strip the subject to shape, material, color and nothing else |
| Extra objects appeared | Context | List the objects you want and add nothing else in frame; do not list the ones you do not want |
| The look is generic or looks like a render | Style | Replace stacked style words with one phrase, or attach a reference image |
| The product floats or the shadow is wrong | Light | State the light direction and ask for a contact shadow on the surface |
| Copy will not fit | Crop | Name the empty area and its side: empty dark space on the right third |
| Every attempt looks the same | Workflow | Change one part per attempt and keep the ones that worked |

Keep the prompts that worked next to the images they made. A sentence that produced a usable hero in October is the starting point for the gift version in December, and a library of forty tested prompts is worth more than any generator of them. For the ad layouts that sit on top of these scenes, [Black Friday ad prompts](/learn/black-friday-ad-prompts-sale-posters) covers typography and headline placement, which follow different rules.

---

Writing AI image generation prompts is closer to briefing a photographer than to casting a spell. The models reward the same clarity a person would: what is in the frame, what surrounds it, where the light comes from, how it is cropped.

Once the formula is a habit, the interesting work moves from the sentence to the choices inside it, which objects, which light, how much room to leave, and those are creative decisions a brand can own and repeat.

## FAQ

### How do you write a good AI image generation prompt?

Write one sentence in four parts: the subject, the setting around it, the style, and one technical detail about light or camera. Name objects and materials rather than moods, say where the light comes from, and describe what you want in the frame instead of what you do not. Then generate, look, and change one part at a time.

### How long should an AI image prompt be?

One sentence of 20 to 40 words covers most product photos. Longer prompts do not add control; they add competing instructions, and the model picks which ones to honor. If a prompt needs a second sentence, it is usually a second image. Put the subject first, because the words at the start of a prompt carry the most weight.

### Why does the AI change my product when I write a prompt?

A text prompt describes the product from words, and words cannot hold a label, a cap shape or an exact color. Tools that anchor generation to a photo of the product keep it accurate and use the prompt only for the scene. If your tool has no product anchor, keep the product description minimal and expect to reject more results.

### Should I use negative prompts?

Describe the outcome you want instead. A model told to avoid blurry text spends attention on text; a model told to produce a clean, empty dark surface on the right produces it. Positive phrasing such as sharp lettering, well-lit scene or empty background works better than a list of exclusions in most current image models.

### What are good AI image prompt examples for product photography?

A matte black marble pedestal with warm gold rim light from the left and the product sharp and centered, for luxury goods. A wet black floor with red and blue neon tubes behind, for tech. A charcoal studio with one softbox and a long shadow, for apparel. Each names objects, a light direction and a crop rather than adjectives.