AI Product Photography: What Works, What Fails, and How to Check Every Image
On this page
AI product photography works best when you start from a real photo of the exact product and let the tool change what is around it, such as the background, shadow and setting. It fails when the tool also changes the product: a different label, the wrong colour, a bent shape. Listing platforms ask for an accurate view of what you sell, so every image needs a check before it goes live.
Plan details below were read from vendor pages on 2026-11-02. We did not run these tools on a product set for this article, so we rank nothing and claim no quality results. The test at the end tells you how to run your own.
What AI product photography does well and badly
The common use is "real product, new scene": you upload a clean packshot and the tool places it in a studio, a kitchen or a flat lay. The risks sit in four places.
| Risk | What it looks like | How to catch it |
|---|---|---|
| Product drift | Label text, logo, buttons or proportions change | Put the original and the result side by side at full zoom |
| Colour shift | The product looks warmer, cooler or more saturated | Compare against a colour swatch or the physical item |
| Scale and physics | The product floats, is too large for the hand holding it, or casts the wrong shadow | Look at contact points and shadow direction |
| Invented parts | Extra straps, wrong number of items, new accessories | Count parts against the product page |
What the rules say
Marketplace rules decide how far you can go. Google's product image guidelines say the main image should give a clear, unobstructed view of the exact item being sold, with each colour or style variant shown in its own image. They also bar promotional text, watermarks and added logos in post-production.
The same page covers AI directly: images made with generative AI must keep the metadata that marks them as AI-generated, such as the IPTC DigitalSourceType tag, and you should not strip it, including from images made with Google's own Product Studio. If your workflow exports through an editor that removes metadata, check that the tag survives.
Plans and prices we could read
| Tool | What its page shows (read 2026-11-02) | Catch |
|---|---|---|
| Pebblely | Lite $9 a month for 30 images, Basic $19 for 200, Pro $39 for 500, with 40+ background themes and custom prompts | Its pricing page does not state commercial-use terms |
| Photoroom | Pro, Max and Ultra plans billed yearly, plus custom Enterprise, with AI Backgrounds, AI Fashion Models and, on higher plans, video | No dollar prices appear in the text we could read, and the page lists some features inconsistently, so check it in your browser |
| Adobe Firefly | US$9.99 a month for 2,000 generative credits up to US$199.99 for 50,000 | Credits are not images; a generation can use more than one credit |
Pebblely's cost per image at full use is simple arithmetic from its page: $19 for 200 images is about $0.095 per generation, and $9 for 30 is $0.30 (pricing page). That is cost per generation, not per usable image. If 1 in 4 results passes your checks, the cost per usable image is four times higher.
Who owns the images
Terms differ. Pebblely's terms, last modified 14 February 2024, say you own the images you generate, but do not address commercial use directly and do not guarantee the images avoid infringing others' rights. Adobe says outputs from its own Firefly models can be used commercially, and that content from partner models is your responsibility to assess. For the wider picture on ownership, see our guide to commercial use of AI-generated images and the copyright explainer.
A 10-product test you can run this week
- Pick 10 products: 3 simple (a box), 4 with printed text, 3 with reflective or transparent parts.
- Use one clean original photo per product, shot on a plain background with even light.
- Run each tool with the same scene brief. Our prompt templates give a starting point.
- Score each result pass or fail on the four risks in the table above. Do not average; one failed label is a fail.
- Record time spent fixing, not only generation time.
- Divide plan price by the number of passes to get cost per usable image.
Then run every passed image through our AI creative QA checklist before upload.
What we have not tested
We did not generate images with any tool named here. We could not read Photoroom's dollar prices. We did not check each marketplace's own image policy beyond Google's. Tool quality changes quickly, so any test result is true for the date you ran it.
Frequently asked questions
Can I sell products using AI-generated images?
Often yes, if the image shows the product accurately and the platform's rules and the tool's terms allow it. Read both, and keep the original photo as evidence of what the product looks like.
Do I have to label AI product images?
Google's guidelines require that AI-generated images keep their AI metadata tag. Disclosure rules for ads are separate; see our guide to AI ad disclosure.
Is a free tool enough?
For a small test, maybe. A free tier is useful for checking drift on your own products before you pay. Check its terms on commercial use first.
