A Practical Text-to-Image Prompt Structure

Product visuals are not art projects. They are shots, and a shot is defined by decisions: what the object is, what it sits on, where the light comes from, how close the camera is, and what color world it lives in. When you write a text prompt, you are making those decisions in words. If you leave them out, the model makes them for you, and the result drifts toward the average of everything it has seen. If you make them explicitly, the output becomes predictable enough to iterate on.

This guide breaks down a repeatable prompt structure for product-style stills, in the order that gives you the most control. The aim is a prompt you can adjust one variable at a time — a first draft you can tune, not a lottery ticket.

A practical note before we start: the structure below is written for the AI text-to-image tool, but the same logic applies to any image workspace that accepts a positive prompt, a negative prompt and a few settings. The exact credit cost is shown in the studio before submission, so check it there rather than trusting a number from any article.

A prompt is a shot brief

Think of the prompt as a brief you would hand someone setting up a tabletop photo. A decent brief covers six things: the object, what it is made of, what it is standing on, where the light comes from, how the camera is positioned, and the color palette. A seventh item — what you do not want — is a short exclusion list, not an essay.

A prompt that lists these in a fixed order is easier to debug. If the glass bottle looks wrong, you adjust the material clause. If the background competes with the object, you adjust surface. This is the whole point of the structure: it turns "regenerate until it looks okay" into "change one clause and check what happened."

Here is the skeleton:

Subject → Material → Surface/background → Light direction and mood → Lens and framing → Palette → Negative list.

Prova questa ricetta

Da testo a immagine AI

Apri lo strumento completo

Caricamento dello studio...

Subject and material

Start with the object and its material, in the same clause. "A ceramic bottle" is not enough; "a matte ceramic bottle with a slight glaze, stopper pushed halfway in" gives the model something to hold onto. Material is what makes surfaces read as real: brushed steel, frosted glass, oiled walnut, ribbed silicone, unglazed clay, powder-coated aluminum, single-ply cardboard with a visible corrugation edge.

Be concrete where the texture matters and leave out detail where it does not. A good rule: describe the surfaces the camera will actually see. If you are shooting the front of a can, the shoulder and lid matter; the back does not. If the object is reflective, say what it reflects — "brushed steel body with soft white reflections from a large source" — because otherwise the model invents a reflection that may not match your scene.

Small inconsistencies in material terms can change the whole image. "Frosted glass" reads soft and diffused; "clear glass with condensation" reads cold and wet. Both are useful, but they are different shots. Pick the one that matches your intent and write it plainly.

Avoid stacking adjectives: three competing finishes on one object will average into something muddy. One material description per part is usually enough.

Surface and background

The surface is the second biggest lever after the subject. It tells the viewer whether the object is premium or utilitarian, and it gives the light something to bounce off.

State the surface and background separately. "Standing on a pale travertine slab, seamless warm-grey background falling into shadow" is a scene. "On a table" is not. Where possible, describe the plane the product rests on, the angle it occupies, and what is behind it — a seamless paper sweep, a dark wall, a window at the frame edge, a soft fabric fall-off.

Keep the background simpler than the product. Busy backgrounds steal attention, and models tend to fill empty space with invented props. If you say "a few scattered leaves near the base" you will usually get more than a few. If you want deliberate styling, be specific: "two dried eucalyptus sprigs, out of focus, bottom right" sets a clearer expectation than "styled with botanicals."

For reflective or translucent products, the surface matters more than usual. A bottle on a mirror will show its own reflection; the same bottle on a paper sweep will not. Decide which one the shot needs before you write the prompt.

Light direction and mood

Light direction is the single most common omission in prompts, and one of the most noticeable when it is missing. The default tends to be a flat, frontal, evenly lit look that reads as catalog filler. Say where the light is and roughly what kind it is.

Useful directions and their effect:

Add one mood word, not four. "Warm mid-morning light" or "cool studio light" is enough. Two or three mood words in a single prompt tend to conflict; the model resolves the conflict by softening everything.

A useful sentence pattern: "lit by a large soft source from the upper left, shadow falling toward the right, soft fall-off into a darker background." This tells the model where the light is, where the shadow goes and how it ends. It is specific without being fussy.

Lens, framing and depth of field

Camera language sets the feel of the shot. For product stills, a few values are usually enough:

• Focal length feel: "35mm look" reads wider and more contextual; "85mm look" reads compressed and product-focused; "100mm macro" is for close detail.
• Framing: full product centered, three-quarter view, top-down flat lay, tight crop on the label, or product entering the frame from one side.
• Depth of field: "shallow depth of field, soft background blur" or "deep focus, everything sharp." These read very differently on the same subject.

The important thing is to choose and state it. If you don't, the model guesses, and you get the same middle-of-the-road angle every time. A prompt that says "three-quarter view, product slightly left of center, shallow depth of field" is already better than most.

Aspect ratio and orientation are settings rather than prompt text, but they interact with framing. Decide your output shape before you write the lens clause; a vertical crop with a top-down flat lay works differently from a horizontal one with a three-quarter view.

A short negative list

Negative prompts work best as a short, focused exclusion list. Long lists of adjectives tend to over-constrain the output and produce flat, defensive results.

Keep to the failures you actually see. For product visuals, the usual suspects are:

• text and watermarks
• extra objects or duplicate products
• distorted reflections
• logo artifacts
• overexposed highlights on reflective surfaces
• cluttered backgrounds
• hands and people, if you don't want them in frame

Write these as comma-separated terms and stop. Five to eight terms is usually enough. If your negative list is longer than your positive prompt, the model is spending more effort avoiding things than making the shot.

One caveat worth stating plainly: negative prompts reduce a behavior, they do not guarantee its absence. Text, in particular, may still appear in small or distorted forms. If a generated image is nearly right but has a stray mark or unwanted object, the product photo editing workflow — mask the area, inpaint, keep the rest — is usually faster than endless regeneration.

Iterating one variable at a time

Here is a working example. Start here:

A matte ceramic bottle with a half-pushed stopper, standing on a pale travertine slab, seamless warm-grey background, large soft light from the upper left with a gentle shadow to the right, 85mm look, three-quarter framing, shallow depth of field, warm neutral palette. Negative: text, watermark, duplicate product, cluttered background, hands.

That will generate something plausible. Now change one thing: swap "85mm look" for "35mm look, slightly wider framing, more surface visible" and regenerate. Compare. Then revert and change the light direction. Then the surface. This is how you find the version that matches your intent, and it is far more efficient than rewriting the whole prompt each time.

A short review checklist before you commit to an image:

• Is the product the clear focal point, or is something else competing?
• Does the material read as the material you specified?
• Is the light direction visible in the shadow and highlights?
• Is the framing what you asked for, or did the model pull back?
• Are there unwanted props, text or duplicate objects?
• Does the palette match the brand or the shot's intent?

If two or more answers are no, change one variable at a time and check again. If you're close but not quite there — a stray shadow, a warped label, an object that should not be there — the editing guide linked above covers the cleanup path. If you're evaluating how many generations a project will need, pricing and the studio's live credit display will tell you the real cost before you run anything; the exact credit cost is shown in the studio before submission.

The structure is not a formula for creativity. It is a way to make each decision visible so that when a shot works, you know why — and when it doesn't, you know which clause to touch next.