How to Create Professional Product Photos With AI
The thing most guides skip: AI product photography still starts with a photograph. These tools don’t invent your product from a text description — you upload a real shot of the bottle, the candle, the jumper, and the AI removes the background, fixes the light and places it in a scene.
Which means the quality of your input decides the quality of your output, and the single biggest improvement available to you is free: twenty minutes of care with a phone and a window.
This guide covers both halves — how to shoot properly, then how to process. There’s a shot list, a lighting setup using things you already own, and the phone settings that matter.
Step 1: Set up a shooting spot that costs nothing
A window, a table and a sheet of card
You don’t need a lightbox. You need soft, directional light and a plain surface, and a window gives you both for free.
Position: put a table beside a window so the light comes from the side, roughly at ninety degrees to where you’ll stand — not from behind you, which flattens everything. A large sheet of white paper or card curved up the wall behind gives you a seamless background with no visible edge where table meets wall.
Light: aim for mid-morning or a bright overcast day. Overcast is genuinely better than direct sun, which creates hard shadows. Turn off overhead lights entirely — mixing daylight with warm bulbs produces colour casts that are a nuisance to correct.
Fill: prop a second sheet of white card on the shadow side, angled to bounce light back. This one change does more for a product photo than any amount of editing, because it opens up the dark side without flattening the shape.
Never use the flash. It produces harsh reflections and a flat, cheap look that no AI tool fully repairs.
Step 2: Work through the six-shot list
Six images cover every listing slot you’ll need
Shoot all six in one session per product. It takes about twenty minutes and gives you a complete set rather than one hero image and a gap where the others should be.
The six-shot list
| Primary | Straight-on, product filling around 85% of the frame, plain white background. This becomes your marketplace primary image. |
| Angle | Three-quarter view from slightly above. Shows depth and form in a way a flat-on shot can’t. |
| Detail | Close on texture, stitching, grain, finish, or whatever justifies your price. Get close — this shot answers “is it well made?” |
| Scale | Next to something familiar, or held in a hand. Prevents the single most common return reason: it arrived smaller than expected. |
| In use | The product doing its job, in context. This is the shot AI scene generation can help with if you can’t stage it. |
| Label | Back, base or packaging — ingredients, specifications, care instructions, dimensions. Skipped constantly, and it removes a whole category of pre-purchase questions. |
Take three or four frames of each and pick the best later. It costs nothing and you’ll almost always find one is sharper than the others.
Step 3: Clean up the shots
Background, brightness, consistency
This is the least glamorous step and the one that most changes how professional a listing looks. Run every image through background removal and basic correction — Photoroom, Canva’s background remover, or Pixelcut all do this in seconds, and free tiers handle it.
What to actually do: remove or replace the background with pure white for your primary, correct the exposure so the product reads clearly, straighten anything slightly off, and crop consistently so the product sits the same size in every image across your catalogue. That consistency is what separates a considered shop from an accidental one.
What not to do: over-brighten until the product loses detail, saturate colours beyond what the item actually is, or smooth away texture. All three cause returns.
Step 4: Generate lifestyle scenes for the secondary slots
Where AI earns its keep
Your primary image needs a plain background. Your secondary images are where products sell, and this is where AI genuinely replaces a styled shoot — placing your real product into a kitchen, a bathroom shelf, a desk, a garden, without hiring anyone.
Upload your cleaned-up cutout, pick a scene or theme, and generate. Theme-based tools like Pebblely are the easiest starting point and have generous free tiers; canvas-based tools give more control at the cost of a learning curve. Our comparison of AI product photo tools covers which suits which catalogue size.
What to check in every generated image: that the shadow falls in the same direction as the light on the product, that reflections make sense, that the scale relative to surrounding objects is plausible, and that any text or logo on the product hasn’t been mangled. Fine text and reflective surfaces are where these tools still fail most often.
Step 5: Check the marketplace rules before you upload
Different channels, different requirements
Major marketplaces generally require a clean white background on the primary image, with no props, no text overlays, no watermarks and no borders — and they’ll reject or suppress listings that break it. Your lifestyle scenes belong in the secondary slots.
Check three things per channel: the primary image requirements, the minimum pixel dimensions (usually larger than you’d guess, because of zoom functionality), and whether AI-generated or substantially altered imagery needs disclosing. That last requirement has been tightening, so verify current rules rather than assuming.
Practical habit: keep one folder of compliant primaries and one of lifestyle secondaries per product, named clearly. It saves rebuilding the set every time you list somewhere new.
Step 6: Write the listing copy from the photos
Make the words match what the images show
Once you have the images, use them to write copy that answers what a buyer is looking at rather than describing the product in the abstract.
PRODUCT: [name and what it is]
MATERIALS / CONTENTS: [specifics]
DIMENSIONS AND WEIGHT: [exact]
PRICE: [amount]
WHO IT’S FOR: [specifically]
WHAT MAKES IT WORTH THE PRICE: [the honest answer]
WHAT IT’S NOT SUITABLE FOR: [be clear]
MY IMAGES SHOW: [list your six shots]
Write:
1. A title under 70 characters, leading with what it is rather than my brand name
2. Five bullet points, each answering a question a buyer would actually have — start with the one most likely to stop them buying
3. A short description under 150 words
4. Suggested alt text for each of my six images
Rules: no superlatives I can’t evidence; include exact dimensions since size is the commonest return reason; state clearly what’s not included; don’t invent features or materials.
Alt text is worth doing properly — it helps accessibility, it helps search, and it takes seconds once you’re already there.
Step 7: Find out whether the new photos actually work
The step almost nobody takes
The premise of everything above is that better photos sell more. That’s well supported in general and unproven for your specific products, so treat it as a hypothesis about your own customers.
The simple method: change images on a subset of products and leave a comparable subset alone, then compare conversion rate rather than revenue, which moves for a dozen unrelated reasons. Give it a few weeks. Some platforms offer built-in image experiments, and one AI photo tool includes A/B testing for listing images specifically.
Watch return rates alongside conversion. An image set that lifts sales while increasing returns is a net loss once reverse logistics are counted — and it’s usually a signal you’ve edited past the honesty line in Step 3.
Before you publish an image set, check
- Product colour matches reality — compare on screen against the actual item.
- Scale is unmistakable from at least one image.
- Shadows and reflections make sense in any generated scene.
- Text and logos on the product are intact, not warped.
- Primary meets the marketplace spec — white background, no props, no overlay.
- Dimensions are stated in the copy, not just implied by a photo.
- You’ve kept an unedited original of every product on file.
Frequently asked questions
Can I skip the photography and generate the product from a description?
No, and you shouldn’t want to. Almost every AI product photography tool worth using works from a real photo of your actual item — it removes the background, corrects the light and places the product in a generated scene, but the product itself is yours. Tools that generate a product purely from text produce something that doesn’t exist, which for a listing is misrepresentation regardless of how good it looks. There’s a narrow exception for concept and mock-up work before a product exists, but that’s design rather than listing photography. The practical upshot is that the twenty minutes with a window and a sheet of card in Step 1 isn’t optional preparation you can automate away — it’s the raw material, and improving it improves every downstream image at no cost.
Which products does AI struggle with?
Three categories, consistently. Reflective items — glassware, polished metal, jewellery — because generated environments produce reflections that don’t match, and the eye notices immediately even when it can’t say why. Anything with fine text or a logo, since these tools frequently warp small lettering, which is a particular problem for cosmetics, supplements and packaged goods. And transparent or translucent products, where background replacement tends to produce edges that look cut out rather than see-through. Fabric drape is a fourth, softer case: AI handles a folded jumper fine and struggles with how a garment hangs. For any of these, do the cleanup steps but shoot the lifestyle context for real, or keep to plain backgrounds and invest in one properly photographed hero image. Test on a couple of products before committing a whole catalogue to a workflow that won’t suit them.
How many images does a listing actually need?
Six is a good working target and covers every slot most channels offer — primary, angle, detail, scale, in-use, label. More than eight rarely adds anything, and buyers seldom scroll that far. What matters far more than count is coverage: each image should answer a different question rather than showing the same view repeatedly, which is the commonest mistake. Four shots of the front from marginally different distances is worse than four shots doing four jobs. If you’re short on time, prioritise in this order: a clean primary, then scale, then detail, then in-use. Scale ranks that high because size confusion is one of the biggest single drivers of returns, and one photo next to a familiar object removes most of it. The label shot is the one people skip and it quietly prevents a lot of pre-purchase questions.
Do I need to disclose that images are AI-generated?
Increasingly, yes on some channels, and it’s worth checking rather than assuming. Several marketplaces and social platforms now expect labelling where imagery has been generated or substantially altered, and those expectations have been tightening rather than relaxing. The distinction generally drawn is between routine correction — cropping, exposure, background removal — and generating content that wasn’t photographed, such as placing a product in a scene that never existed. The second is where disclosure questions arise. Practical approach: check the current policy for every channel you sell on before generating a catalogue’s worth of scenes, keep your primary images as straightforward corrected photographs, and never generate an image implying a variant, feature, size or setting you can’t deliver. That last point matters more than the labelling question, because misrepresentation causes returns and complaints whether or not it was disclosed.
The bottom line
AI hasn’t removed the photography, it’s removed the studio. So spend your effort where it compounds: a window, a sheet of white card as a bounce, the 1x lens, a wiped lens, and six shots per product that each answer a different buyer question. Then let AI handle background removal, correction and the lifestyle scenes your primary image can’t be. Keep your primaries compliant with each marketplace, keep the product honest — an image that flatters past reality returns as a refund — and actually measure conversion afterwards, because that’s the only number that tells you any of it worked.
Tool capabilities and free-tier limits change frequently — check current terms, and confirm commercial-use rights before publishing. Marketplace image specifications and AI disclosure requirements differ by platform and are being tightened: verify the rules for every channel you sell on. Nothing here is legal advice; misrepresenting a product in imagery may engage consumer protection law.