# AI product photography without a studio — materials, labels, sets

> Shoot a product without a studio — how to describe materials and labels, use a reference photo, and save the product as an Element for a whole catalogue.

Published 2026-09-04 · eroq.ai — canonical: https://eroq.ai/blog/ai-product-photography-guide


A product shot is three things: an object, a surface it sits on, and a light that makes the material read. A generative model is good at two of them immediately and good at the third once you stop describing your product by its category and start describing what it is made of. This is the working method for shooting a catalogue in [the Image studio](/studio/image), including the two things it still cannot do for you.

## Describe the material, not the category

"A bottle of serum on a white background" gives you a stock photo of somebody else's serum. The model has no idea which object you mean, so it renders the average of every product that has ever been called a serum.

Name the material, the finish, the closure, the color of the glass, the shape of the shoulder. The prefilled object Elements in the studio are written exactly this way, and they are a good template to copy:

- *a small amber glass perfume bottle with a brushed gold cap and a cream minimalist label, catching soft studio reflections*
- *a frosted glass skincare serum bottle with a dropper cap, fresh water droplets, clean clinical-beauty styling*
- *a crisp white leather sneaker with a gum sole and clean stitching, floating hero-product angle*

Every one names a material first — amber glass, frosted glass, white leather — then how light behaves on it. That ordering is not decoration. Material plus light is what separates a photograph from a render, and it is the same vocabulary the [photoreal prompts guide](/blog/photoreal-ai-image-prompts) uses for faces and streets.

## Light and surface do the rest

Three setups cover most of a catalogue, and each one is a phrase rather than a paragraph.

- **Soft cross light on a seamless background.** Two soft sources at an angle, no hard shadow. The clean commercial default for skincare, cosmetics, electronics — anything that should look clinical.
- **Hard studio flash on concrete or stone.** One direction, a crisp shadow, dust in the beam. Sneakers, watches, tools, spirits. It looks expensive because the shadow has an edge.
- **Window light on a lived-in surface.** Linen, oak, marble, a folded towel. Food, candles, ceramics, anything sold on warmth rather than precision.

Say which one you want, and say what the object is standing on. "A white void" and "a slab of raw slate" are two different price points for the same bottle.

## The label problem, stated plainly

Here is the limit you need to plan around: the model will render something that looks like a label, and it will not be your label. Fine print comes out as texture that resembles typography, and an exact logo — your logo, with its real letterforms and spacing — is not something a diffusion model reproduces reliably even with a reference photo in hand.

Three ways to work with that rather than against it:

1. **Compose for a plain label.** Ask for "a cream minimalist label, no text" and put "text, lettering, watermark, logo" in the negative prompt. You get a clean surface, and your real artwork goes on in your editor. This is the honest path for anything that will be seen at full size.
2. **Turn the label away.** A three-quarter angle with the front face rotated out of view makes an attractive hero shot and sidesteps the issue entirely.
3. **Shoot the context, not the packshot.** Hands, use, atmosphere, the product small in a larger scene. At that scale nobody is reading the label anyway.

The same rule applies to screens, price tags and signs inside your frame: rendered text is a texture, never information. The [negative prompt](/glossary/negative-prompt) entry has more on excluding it.

## Hand the model the real object

Descriptions get you a plausible product. A reference photo gets you *your* product. Attach a photo of the actual object and the render starts from that picture rather than inventing a new one — the shape, the proportions and the color come along.

One clean photo beats three messy ones. Front-lit, sharp, filling the frame, on a plain surface, no hands. Attaching exactly one reference puts the render into edit mode, which is also why your format setting stops applying in that case: the source photo already has a shape, and it wins. There is a whole piece on what holds and what changes in the [image-to-image editing guide](/blog/image-to-image-editing-guide), and [reference image](/glossary/reference-image) has the short definition.

## Save the product as an Element

The real unlock for a catalogue is that you should only describe your product once. An Element is a reusable reference — a place, an object or a style — with a name, a written description and photos. Save the product as an object Element and you can @mention it in any image or video prompt afterwards: its description folds into the prompt and its photo joins the references automatically.

That is what keeps a set coherent. Twenty renders that each describe "the amber bottle" in slightly different words will give you twenty slightly different bottles. Twenty renders that @mention the same Element give you one bottle in twenty situations. You can keep up to 50 Elements, and a second Element of kind *style* — "clean beauty editorial, soft diffused light, pastel seamless background, glossy magazine finish" — keeps the whole campaign on one look. The full mechanics are in [how to use Elements](/blog/how-to-use-elements-places-objects-styles).

## A prompt you can run today

Pick the format before you write: 1:1 for feeds and grids, 3:4 for tall packaging, 4:3 for editorial stills, 16:9 for a page header. Then one flowing paragraph, material first.

> Hero shot of a small amber glass perfume bottle with a brushed gold cap standing on a slab of raw slate, hard studio flash from the upper left with one crisp shadow falling right, fine dust suspended in the beam, deep gradient background, condensation-free glass, precise and expensive.

Render a batch of four and you have four lighting variations of the same hero. Change one clause at a time after that — the surface, then the light direction, then the background — because a prompt that changes everywhere tells you nothing about what worked.

## What a catalogue actually costs

Every image is 10 credits on every model, at any format, so the arithmetic is simple. A batch of four is 40. Six angles, each explored in a batch of four, is 240 credits. The $10 starter pack is 1,000 credits — a hundred finished stills — and new accounts land with 50, enough to shoot one product before deciding anything.

When the same product needs to move as well as sit still, the object Element carries straight over to the video tool, and the three-shot commercial recipe lives in [AI product videos for ecommerce](/blog/ai-product-videos-for-ecommerce) and on the [product videos](/use-cases/product-videos) page.

## FAQ

### Can the model put my real logo on the product?

No, not reliably. Rendered lettering comes out as convincing texture rather than your actual artwork, so plan for a plain label and composite the real one in your editor. Turning the label away from camera, or shooting the product in context rather than as a packshot, avoids the problem entirely.

### How do I keep the same product looking the same across a catalogue?

Save it once as an object Element with a written description and a photo, then @mention that Element in every prompt. The description folds into the prompt and the photo joins the references, so you are re-using one definition instead of re-describing the object each time.

### Do I need a photo of the product, or is a description enough?

A description is enough when the product is generic — a plain white mug, an unbranded candle. When it is a specific object with its own shape and finish, attach a photo. One clean, front-lit shot on a plain surface does more than any amount of adjectives.

### What resolution do product shots come out at?

At the native size of the format you pick, from 1024 × 1024 for a square up to 1344 × 768 for widescreen. There is no upscaler, so choose the shape you need before rendering rather than planning to enlarge afterwards.

Describe the material, attach the object, save it as an Element — then [open the Image studio](/studio/image) and shoot the first angle.
