AI Product Photography: White Background to Branded Scene

Product photography AI tools do two jobs well: they cut a product out of a messy photo and place it on a clean background, and they build a scene around that cutout from a text description. What they cannot do reliably is reproduce small text, seams, ports and caps from scratch, so the product itself should always come from a photograph.

When does using AI for product photography pay off?

• One product, many contexts: a serum bottle on a bathroom shelf for the gallery, on stone for an ad, next to a travel bag for social, all from one approved source photo.
• Seasonal refreshes: the same catalog re-shot for a holiday palette without a studio day.
• Marketplace hygiene: turning supplier photos with mixed backgrounds into a consistent white-background set.
• Pre-launch: showing a product in use before physical samples exist, clearly labelled as a render internally.

What can AI not change in a product image?

Anything a buyer relies on: shape, proportions, colour, label text, certification marks, included accessories and quantity. Shopify's ecommerce photography guide notes that the plain white-background style is the one shoppers expect on marketplaces such as Amazon, with lifestyle shots as the creative layer on top, per Shopify (2026). Keep the factual image factual and let the generated scene carry the mood.

The whole route uses three tools in the image workbench and one optional model. Every run below was priced on 2026-09-06 with a promotional discount showing on the button; read the button before you generate because prices change.

1. 1

   Shoot the product on a phone

   Daylight from a window, product on a plain surface, camera at the product's mid-height, no zoom. Take a front view and a three-quarter view. Sharp edges and readable packaging matter more than the background.

2. 2

   Make the white-background image

   Open the AI product image generator, upload the photo and type the surface you want, for example 'white studio background with soft contact shadow'. Choose 1:1 for marketplaces, then 2K or 4K output. The button quoted 13 credits per run on 2026-09-06.

3. 3

   Build the branded scene

   Take the same photo into the AI background changer. It accepts a subject photo, an optional background photo and a prompt, and keeps the main subject intact while replacing everything else. Describe surface, light direction and one or two props.

4. 4

   Upscale before export

   Run the approved image through the AI photo upscaler so zoom views and 1500-pixel marketplace crops stay sharp. Upscale last, after any edits, so you only pay for the final.

5. 5

   Export one file per channel

   Square for marketplaces, 4:5 or 9:16 for social, 16:9 for email banners. Each is a separate composition, not a crop of the same one.

6. 6

   Compare against the source

   Put the output next to the original photo at 100 percent. Check label spelling, cap and handle position, colour and the shadow direction against the light you described.

Marketplaces decide what a generated image may be used for. Google Merchant Center requires product images of at least 500 by 500 pixels from January 31, 2027, recommends around 1500 by 1500 pixels or larger, and rejects files over 64 megapixels or 16 MB, per Google Merchant Center Help (2026). Amazon keeps a dedicated section of MAIN image requirements and, since our last check, a section on AI-generated people disclosure in its Product image guide (2026).

Two rules travel across every channel. Do not add text, badges or watermarks to the image itself, and do not let a generated prop imply a feature, ingredient or bundle the product does not include. The FTC's Endorsement Guides FAQ states that endorsements must be honest and not misleading, per the FTC (2026), and a scene that stages a fake customer is an endorsement.

eBay publishes its own picture policy in the Seller Center; read it in a logged-in browser, because the public URL answered our 2026-09-06 request with a bot check.

Good AI prompts for product photography read like an art director's brief: what stays fixed, then surface, light, props, camera and framing. Exclusions go last. The formula below is what we paste into the background changer and, with a reference image attached, into text-to-image.

Keep the supplied product unchanged: shape, proportions, colours, logo, label text. Place it [position] on [surface] in [environment]. Light: [direction and quality]. Camera: [angle and height]. Props: [one or two, secondary]. Palette: [two colours]. Composition: [ratio], space on [side] for copy. No extra products, hands, new text or watermarks.

Which examples work per category?

• Skincare: 'Upright on pale limestone in a warm-white bathroom, soft window light from the left, subtle contact shadow, one folded linen towel out of focus, beige palette, vertical, clear space above.'
• Food and drink: 'Coffee bag on dark walnut in a small kitchen, warm side light, shallow depth of field, a ceramic cup and a few beans, square, space top-left.'
• Electronics: 'Headphones on a light-grey plinth, seamless neutral backdrop, large diffused light upper-right, crisp edges, front three-quarter view, wide, no props.'
• Outdoor gear: 'Insulated bottle on dry granite at sunrise, golden backlight from behind-left, distant blurred ridge, green and amber palette, vertical, headline space at top.'

For scenes that need the product re-rendered at a new angle, attach the photo as a reference in GPT Image 2, which takes text plus reference images and outputs up to 2K per its model page; it was also the model behind 127 of 181 external generations in LazyKiwi's August 2026 workbench data. When you need several references at once, for example three angles of the same bottle, WAN 2.7 Image Pro reads up to nine images in one run. Re-rendered products still need the label checked against the source.

AI product photography for ecommerce starts from the destination. Square is the default for a reason: in LazyKiwi's workbench data (August 2026), 1:1 was the ratio on 124 of 143 external image jobs. Each channel wants a different ratio, a different amount of scene and a different tolerance for text, so generate per channel rather than cropping one hero image four ways.

Build a small scene system per product line: one neutral, one in-use, one campaign. Reuse the same light direction, surfaces and palette across the line so a set of generated images reads as one shoot. For a product page hero that moves, the product reveal video template turns the approved still into a short clip without a second prompt.

Searches for AI product photography free usually mean no card and no watermark. New accounts land on the LazyKiwi Free plan, 40 credits a month and nothing to pay, per the pricing page. At the 13-credit price the product image generator and background changer showed on 2026-09-06, that covers a handful of runs a month; the paid plans list no-watermark exports and more credits.

Pebblely lists its Lite plan at US$9 a month for 30 images and Basic at US$19 for 200, per Pebblely pricing (2026). Photoroom's pricing page offers a free start and a free trial of Pro, which lists 8,000 AI credits and 1,000 exports, per Photoroom pricing (2026). Both are dedicated product tools; LazyKiwi's advantage is that the same credits also run the image models and video templates.

A product photo generator fails in predictable ways, and most of them are visible only at full size next to the source photo. This is the review list we run before anything is uploaded.

• Changed shape or label: a redrawn cap, a moved button, a misspelled ingredient. Fix: keep the product pixels from the photo; re-render only when a new angle is essential, then compare.
• Wrong scale: a mug the size of a bucket next to a generated chair. Fix: name a familiar prop at true scale in the prompt.
• Fake reflections and floating products: a shadow that points the wrong way or no contact shadow at all. Fix: describe the light direction that matches the source photo.
• Banned backgrounds on MAIN images: a lifestyle scene uploaded as the primary marketplace image. Fix: white MAIN, scenes as gallery.
• Implied claims: generated fruit next to a supplement, a second item that looks bundled. Fix: props that cannot be read as ingredients or contents.
• Text baked into the image: prices, badges, watermarks. Fix: add text in the channel's own tools.
• Over-glossy everything: plastic-looking metal and glass. Fix: name the material and 'matte' or 'satin' where true.

A beautiful image that changes the product is a returns problem, not a marketing asset.

— LazyKiwi creator playbook