A seller can use AI to plan, draft and finish listing infographics and lifestyle images as long as the main image stays a true photograph of the product on a pure white background, and every number or claim on a secondary image traces back to a document you hold. Use AI for layout and copy, never to invent product detail.
The two habits to break are letting an image generator redraw the product so that it no longer matches what ships, and letting a language model write feature claims nobody checked. The stakes are concrete: Amazon's image guide says images that break its requirements may be removed, and a listing with no compliant main image may be taken out of search until one is provided.
What Amazon's own image page actually says
Rules change, so read the current page before you brief anyone. As fetched on 5 November 2026, Amazon's product image guide in Seller Central sets these standards for the main image: a realistic, professional image that represents the real scale, quantity and colour of the product; the product filling 85% of the frame; the whole product visible; the product shown once, with only the accessories that are included; and a pure white background (RGB 255, 255, 255). Only a limited set of product types may use a lifestyle image as the main image, and then with no text or extra logos.
The page also lists rules for every image in the listing. Images must accurately represent the product, match the title, and carry no images of customer reviews, no five-star imagery, no claims such as "free shipping" and no seller-specific information. They must not use Amazon logos or trademarks, or badges such as "Best seller" or "Amazon's Choice". Technical limits: JPEG, TIFF, PNG or non-animated GIF files, at least 500 pixels on the longest side to upload, 10,000 at most, and 1,000 or more to enable zoom.
Two practical points follow. First, the secondary slots are where infographics live. The page's "no text or pricing details" line is marked as applying to swatch images, and the few product types allowed a lifestyle main image must keep it free of text and extra logos; for text on other images, the category style guide for your product type decides what is sensible, so check it. Second, uploading an image does not guarantee it is shown. Amazon selects and arranges images, and can modify them, so a compliant image is a necessary step, not a guarantee.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowWhich job goes to AI and which stays with a camera
| Image slot | AI can do | AI must not do |
|---|---|---|
| Main image | Remove a cluttered background, even out the white, fix a stray shadow or dust | Redraw the product, change its colour, add or remove parts, enlarge it beyond its real proportion |
| Infographic | Suggest the five questions to answer, draft short labels, propose layouts | Produce figures, certifications or comparisons you have no document for |
| Lifestyle scene | Generate a background or room around a real cut-out of your product | Show the product doing something it cannot do, or at the wrong size |
| Size and dimension image | Place callout lines on your real measurements | Estimate dimensions from a picture |
| Packaging or what's-in-the-box | Arrange your real items neatly | Show items that are not in the box |
The pattern is simple. AI is good at the space around the product and at words on a template. It is unreliable at the product itself, because a generator treats a product as a pattern to approximate, not an object to preserve. If a buyer opens the box and the clasp, the stitching or the label differs from the picture, you have a return and possibly a "not as described" complaint.
A worked example: a solar-powered garden light, start to finish
Say a small seller lists a stake-mounted solar garden light, sold in packs of four. The seller has six real photographs from a phone and a light tent, a spec sheet from the supplier and a short list of buyer questions gathered from competitor listings and the seller's own returns. The plan is one main image and six secondary images. The illustrative numbers below show where the time goes.
- Brief (20 minutes). Paste the spec sheet and the buyer questions into a chat assistant and ask for a seven-image plan. The prompt in the next section does this.
- Main image (15 minutes). Take the best photograph, remove the background in an editor with an AI background-removal tool, set the canvas to 2,000 x 2,000 pixels, fill it with RGB 255, 255, 255, and size the product so it fills about 85% of the frame. Check the corners with the eyedropper: they must read exactly 255 across all three channels.
- Infographic one: what's in the pack (20 minutes). Four lights, four stakes, one spare battery, nothing else. Layout from the assistant's suggestion, drawn in a design tool with real photographs.
- Infographic two: how long it runs (30 minutes). The assistant drafts a label. The seller then checks it against the spec sheet and finds the draft says "up to 12 hours" while the sheet says 8 to 10 hours after a full charge in direct sun. The line becomes "8 to 10 hours on a full sunny day's charge".
- Lifestyle scene (25 minutes). The real cut-out of the light is placed on a generated garden path background, scaled against a known object. The seller checks that the stake height in the image roughly matches the 40 cm on the spec sheet.
- Dimensions image (15 minutes). Callout lines on the real measurements, no AI-estimated sizes.
- Compliance pass (25 minutes). A last read of each image against the list in the section after next.
About two and a half hours for seven images in this illustration. Notice that the most valuable minutes were the ones spent catching "12 hours" before it went live.
The brief that turns a spec sheet into an image plan
Do not ask the assistant for "an infographic". Give it your documents and ask for a plan you can accept or cut.
You are helping me plan Amazon listing images for one product.
Use ONLY the facts in the spec sheet and buyer questions below.
If a fact is missing, write "NEEDS SOURCE" instead of guessing.
Output a table with: image number, slot (main, infographic, lifestyle,
dimensions, what's in the box), the one buyer question it answers,
the exact text to appear on the image (max 12 words per label),
and the document each fact came from.
Rules from Amazon's image guide that apply: no review quotes or
star graphics, no "free shipping" or other offers, no Amazon logos
or badges such as "Best seller", no claims I cannot prove.
SPEC SHEET:
[paste]
BUYER QUESTIONS:
[paste]A reply in the right shape looks like this (illustrative): image 2, infographic, answers "How long does it stay lit?", text "8 to 10 hours on a full charge", source "Spec sheet, runtime row". Image 4, lifestyle, answers "Will it work on a shaded path?", text "NEEDS SOURCE". That last line is the system working: the assistant admitted it had no basis, and you either test the product in shade or drop the image.
If you have not collected buyer questions yet, do that first. Review analysis with AI gives you the themes buyers keep raising, which are better infographic topics than your own guesses.
Keeping the main image honest
The main image is the one Amazon is strictest about, and where AI tools tempt you most, because a generator will happily give you a glossier product than the one you ship. Work from a real photograph and use AI only for these jobs.
- Background removal and white fill. Fine, then verify with the eyedropper. A "nearly white" background of 250 or 252 fails the RGB 255 rule even though it looks white on screen.
- Dust, scratches and stray reflections. Fine, as long as you remove blemishes on the photograph, not features of the product. Removing a seam, a screw or a weld mark changes what the buyer receives.
- Shadow. A soft contact shadow is often accepted but changes how the white reads. Check the category style guide before adding one.
- Colour. Match the swatch against the physical product under daylight, not against the generator's output.
A realistic mistake: a seller of powder-coated steel brackets uses an AI "enhance" filter, which smooths the texture and lightens the grey. The photograph looks cleaner, but buyers receive a darker, more textured product and the reviews say "colour doesn't match". The fix was to turn the filter off and correct only exposure.
If you want the detail on what Amazon accepts when AI is involved, the separate tutorial on AI-generated product images and what passes review goes through it.
Infographic text: short labels, checked numbers
An infographic is a picture with a job: answer a buyer question in two seconds on a phone screen. That suggests a tight set of rules for the text itself.
- One question per image. If an image needs a paragraph, it is two images.
- Labels of 12 words or fewer, in type large enough to read when the image is shown at thumbnail size.
- Numbers only from a document: spec sheet, test report, your own measurements. Write the source in the plan, not on the image.
- No superlatives you cannot prove ("the strongest", "number one"), and none of Amazon's reserved badges.
- Comparison tables only against your own previous model or a generic "ordinary" version, never a named competitor's product.
- No offers, no price, no shipping promises, no review quotes.
Before and after (illustrative). Draft from the assistant: "Ultra-bright, weatherproof, lasts all night, premium quality". Fixed version: "IP44 rated", "8 to 10 hours on a full charge", "Stake height 40 cm". The first set is adjectives nobody can verify. The second is three facts, each tied to a line on the spec sheet. Note that an IP (ingress protection) rating is an IEC code whose first digit grades protection against solid objects and whose second grades protection against water. It is a claim about a tested rating, so only use it if the supplier's test report states it, and quote it exactly.
Lifestyle images without overselling the product
A lifestyle image tells a buyer where and how the product fits into their life. It can sit in the secondary slots in most categories. There are two sensible ways to make one with AI.
Composite: real product, generated scene
Cut your product out of a real photograph, then ask an image tool to generate the scene around it (a patio table, a workbench, a hallway shelf), and place the cut-out in the scene yourself. The product stays real, the scene is imaginary but plausible. This is the safest route.
Fully generated: use rarely
Letting a model draw the product in the scene gives you a product that looks similar, not identical. For anything with text, logos, buttons, ports or fine detail, you will see invented features. If you do use a fully generated image, compare it to your product part by part, and drop it if it shows anything you do not sell.
Three checks for any scene. Scale: is the product the right size next to everyday objects? Function: is it shown being used the way it is designed to be used? Contents: does the scene contain items that look like part of the sale but are not (extra stakes, a cable, a second unit)? Amazon's main-image page specifically warns against props that could confuse the customer, and the same sense applies to secondary images because the listing must match the title.
If your scene includes a photorealistic person who was generated entirely by AI, Amazon's guide asks you to add the metadata keyword contains-synthetic-performer in the image file's XMP subject field before upload. It does not apply to real people edited with AI, or to people who are not photorealistic. Avoiding AI-generated people altogether sidesteps the question.
Eight things to look for before you upload
Run every image through this list and write the answer in a spreadsheet beside the file name. It takes about three minutes per image once you are used to it.
| Check | How to verify | Typical failure |
|---|---|---|
| Matches the title | Read the title, then the image | Image shows a six-pack, title says four |
| Real scale, colour, quantity | Hold the product next to the screen | Generated colour looks richer than the stock |
| White background (main) | Eyedropper on all four corners reads 255, 255, 255 | Soft grey vignette left by the editor |
| Product fills about 85% (main) | Measure the bounding box against the canvas | Lots of white space around a small product |
| No reserved badges or logos | Search the image for "Best seller", "Choice", "Prime" | A "Top rated" ribbon added by a template |
| No offers or seller details | Read every word on the image | "Free gift" or a website address |
| Every figure sourced | Match each number to a line in the plan | "Up to 12 hours" when the sheet says 8 to 10 |
| Size and file type | Check pixels on the longest side, format, RGB | 1,200 px image stretched from 600 |
Build a reusable template so the second product takes half the time
Once the first listing is done, save the plan prompt, the canvas template at 2,000 x 2,000 pixels, a style sheet (fonts, two or three colours, icon style) and the compliance table. A seller with a dozen near-identical products, such as sizes or colours of one design, can reuse the layouts and swap in the product-specific photographs and facts.
Two sets of images that share layout but not facts is the point. A repeated layout with a repeated claim is how an error spreads to every listing, so the facts column in your plan is per product, not per template.
If you sell the same product on several marketplaces, the dimensions and overlays may need to change per site, because each has its own rules. The comparison in multichannel listing across Amazon, eBay and Etsy covers that. If you translate overlay text, translating a listing with AI shows how to keep claims intact when words change.
Filling the bullets and the images with the same facts
Images and bullets should tell the same story. A bullet that says "8 to 10 hours" and an image that says "12 hours" is a credibility problem for buyers and a reason for a customer complaint. Keep a single fact sheet per product: one row per claim, its value, its source document and where it appears (title, bullet, image number). When a figure changes, you update the sheet and see every place it needs to change.
That fact sheet is also the input to your listing copy. The companion tutorial on AI tools for Amazon listing copy shows how to keep bullets and search terms consistent with it.
Five realistic ways this goes wrong
- The generator adds a feature. A tool bag in a scene gains an extra pocket the real bag lacks. Fix: composite the real cut-out, or compare the generated product with the real one at 200% zoom.
- The infographic quotes a rival. A comparison column uses a named competitor's specification taken from memory. Fix: remove named competitors, and compare against your own earlier model.
- The text is correct in English and wrong in translation. A translated overlay turns "water-resistant" into a word that means "waterproof". Fix: have a fluent reader check every overlay, and keep claim wording in the fact sheet.
- The white background is not white. An export profile shifts the white to 254. Fix: export in sRGB, check the corners after export, not before.
- An old image outlives a product change. The supplier changes the battery capacity and the infographic still says the old figure. Fix: the fact sheet records the date each source was last checked, and a spec change triggers a review of every image that uses it.
Checking that the images are working after launch
Images are a hypothesis about what buyers need. After a few weeks, look at three signals. First, returns and customer questions: if "does it come with stakes?" keeps arriving, the what's-in-the-box image is not clear enough. Second, the sessions and sales figures in Seller Central's reports for the listing, compared before and after an image change, changing one thing at a time where you can. Third, reviews that mention the product not matching its pictures, which is the signal that something on an image overpromised.
Keep the changes small and logged. Swap one secondary image, note the date, and compare a few weeks either side. A test with no log teaches you nothing. If your brand is enrolled in Brand Registry, you have a Professional selling account and the listing gets enough traffic, Amazon's Manage Your Experiments tool can run a proper A/B test on product images instead.
The wider point from how small Amazon sellers use AI for listings and reviews applies here: AI speeds up the drafting, and a person who knows the product decides what the page is allowed to say. If you would like help setting up a brief and a checking routine for your own catalogue, that is what my AI implementation consultation covers, and you can reach me through the contact page.
Questions sellers ask before uploading AI images
Can I use an AI-generated image as my Amazon main image?
Only if it is a realistic, professional image that accurately shows the real product, with its true scale, colour and quantity, on a pure white background. A generated image of a product you have not photographed risks showing details that do not exist. Start from a real photograph and let AI clean the edges.
Does Amazon require a label on AI-generated people?
Amazon's product image guide says some places may require disclosure of photorealistic AI-generated people, and tells sellers to add the metadata keyword contains-synthetic-performer to such images before upload. It does not apply to real people edited with AI or to people who are not photorealistic.
Are 1,000 pixels a requirement?
No. Images need at least 500 pixels on the longest side to upload, and 1,000 or more on the longest side enables the zoom function, which Amazon recommends. Do not artificially enlarge small images. Exporting at 2,000 pixels on the longest side keeps you clear of both the minimum and the zoom threshold.
Who checks the numbers and claims on my infographic?
You do. AI writes confident figures that may not match your spec sheet. Every number, certification and comparison on an image should be traced to a document you hold, and anything you cannot prove should come off the image before it goes live.
Further reads
- How to Create Product Photos With AI on a Small Budget — Cheap ways to make the photographs these images start from.
- AI Product Photos vs a Photographer: Cost and Quality Compared — Decide which shots still need a camera and a person.
- How to Write Product Descriptions With AI That Actually Sell — Write the copy that sits beside your images.
- Online Shop AI Mistakes That Hurt Trust and Conversions — Mistakes that cost shoppers' trust, images included.
- Can a Small Online Shop Use AI to Compete With Bigger Brands? — Where a small seller can use AI against bigger brands.
- Amazon FBA vs FBM Cost: Model Your Own Products With AI — A per-unit spreadsheet model for FBA versus FBM, built with ChatGPT or Claude, with two worked products and the break-even sums.
- Amazon Seller Virtual Assistant or AI: Which Tasks Go to Whom — A task-by-task split for Amazon sellers: what a virtual assistant should do, what AI should draft, what stays with you, and what each costs.
- Is AI Amazon PPC Software Worth It? Break-Even for Small Sellers — A break-even sum for AI Amazon PPC software, with fee-to-spend tables, a manual-review baseline and a one-campaign test plan.
- Walmart Marketplace vs Amazon: Which Should a Small Seller Add Next? — Approval, fees, fulfilment, competition and ads compared, with a decision rule by product type for the seller who has one channel and wants a second.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Amazon Seller Central product image guide and product title requirements (fetched 5 November 2026), Amazon's note on tagging AI-generated people, Amazon's Manage Your Experiments page and the IEC page on IP ratings (all read 5 November 2026).