Dropshipping Product Photos: Replace Supplier Images Safely

Yes, you can replace supplier images with AI-made or AI-edited product photos, but only when you have the right to use whatever the AI started from, the picture shows the real product accurately and you follow each sales channel's rules on AI imagery. The safest route is to photograph a sample and use AI to improve it.

The temptation runs the other way: type a product name into an image generator and publish whatever looks good. That works until a buyer opens a parcel containing something a different shape, shade or size from the picture. In dropshipping, where you never touch the goods, the picture is your product, and the return is your cost.

Follow me on Instagram@sagnikteaches

Why not just keep the supplier's photos?

Supplier photos are free and fast, and that is the problem. Every other store selling the same item uses the same ones, so they tell shoppers and search engines nothing unique about yours. They may also be small, watermarked, low-quality or taken for a listing of a slightly different version. And they may not be yours to use at all.

Connect on LinkedInSagnik Bhattacharya

Rights are the part owners skip. A supplier's product page can show photographs taken by a studio, copied from a brand or lifted from a competitor. Having a link to a picture does not give you a licence to use it. Some suppliers say in their terms that they allow partners to use their images; some do not say anything; some images are plainly someone else's. Before building a store on any set, ask the supplier in writing: "May I use these images on my own store and in adverts, commercially, for as long as I sell this product?" Keep the reply.

Subscribe on YouTube@sagnikteaches

None of this is legal advice. Image licensing differs by place and by contract, and if a brand or photographer complains, the cost lands on the store owner. Ask a qualified adviser if you are unsure, and prefer pictures you made yourself or commissioned.

The four sources of a product image, from safest to riskiest

The pattern is clear. AI is safest as an editor of a true photograph and riskiest as an inventor of one. Everything below follows from that.

What "accurate to the real product" means in practice

Accuracy has five parts, and image generators get each of them wrong in different ways:

• Shape and construction. Straps, buckles, zips, handles and joins are the details an image model most often invents or simplifies.
• Colour. Screens vary, but a "sage green" jumper that arrives olive is a classic return. A generated image tends towards attractive colours, not true ones.
• Size and scale. A generated hand holding a "small" pouch can make it look twice the real size.
• Features and ports. Extra buttons, an extra cable or a charging port that is not there.
• What is included. Props in a lifestyle photo (the plant, the book, the cushions) read as part of the deal unless you say they are not.

Google's Merchant Center guidance puts the standard plainly for its listings: the main image must show the exact item being sold, with placeholders, shared stand-in photos and illustrations that do not depict the physical product to be avoided. It also tells merchants to avoid stock-photo watermarks, copyright symbols, promotional text and added logos on the main image. Whatever your channel, treat that as a baseline.

A safe workflow: sample, shoot, edit, check

1. Order a sample. One unit of each product before you scale ads. It is a cost of testing, and you also learn the real delivery time and the quality.
2. Shoot the true product. A phone, a plain white sheet or a large sheet of paper, window light and a tripod or a stack of books. Take front, back, side, detail close-ups and one shot with a hand or coin for scale.
3. Edit with AI. Use AI to remove a messy background, even out the lighting, crop to your platform's ratio and, if you like, place the product on a styled scene. The product pixels stay yours.
4. Check against the sample. Put the finished image next to the real item. Walk through the five-point list above. Fix or discard anything that fails.
5. Record where each image came from. A sheet with columns for product, image file, source (own photo, licensed, AI edit), tool used and date. This is your evidence if a platform or a buyer queries an image.

The edit step is where AI earns its keep. Tools such as the image editor built into Shopify's admin offer background removal and generation. Check the product's current feature list in the Shopify help centre: at the time of the last check, its file editor "Generate" option produced images of about one megapixel with an invisible watermark, which is fine for a thumbnail and not enough for a full-width hero on a large screen. General-purpose assistants also take image uploads (ChatGPT accepts images of up to 20 MB each), and can edit a supplied photo on the instruction you give.

Prompts that edit instead of invent

The wording of a prompt decides whether the tool touches only the background or redraws the product. A prompt that says "a stylish water bottle on a kitchen counter" asks for an invention. This one asks for an edit.

Use the attached photo of my product exactly as it is. Do not change the product's
shape, colour, proportions, logo, text, stitching, buttons or any detail on it.
Replace only the background with a plain light-grey studio surface and soft shadow
under the product. Keep the product the same size relative to the frame.
Output a square image.

A lifestyle version adds a scene while locking the product:

Place the product from the attached photo, unchanged, on a wooden desk beside a
laptop and a mug, in soft morning window light. Do not alter the product itself.
Do not add any extra accessories to the product. The mug and laptop are props and
are not included with the product.

Typical result (illustrative): the background and the props come out well; on the second pass the product's fastener pull has drifted slightly from silver to black. That is the kind of subtle drift you only catch by comparing the output with the real item or the original photo. Zoom to 200% and check logos, text, buttons, stitching and colour against the original. If anything moved, regenerate or fix by hand.

The same approach lets you build a set of consistent images for the gallery: front, three-quarter, detail and in-use. Keep the settings identical (background, lighting, crop) across the set so the product page looks like one shoot rather than six sources.

Returns: the number that decides whether AI images pay

An attractive image that sells more units and returns more units may leave you worse off. Do the sum with illustrative figures. Say a product sells for $29.00 and the supplier charges $14.00 including shipping. A refunded order costs you the full $29 you hand back, and because a dropshipped item is rarely worth shipping back to the supplier, the $14 you paid for it is gone too: you recover nothing.

With real photos you get 100 orders and a return rate of 6%. With glossy AI images that do not quite match, orders rise to 125 but returns hit 14%.

Twenty-five extra orders leave the glossy version only about $41 ahead, and that is before card fees, support time, chargebacks and reviews. Add a fee of 2.9% plus 30 cents on every order (the standard published rate for Shopify Basic's online card payments; check your own plan), which is about $1.14 on a $29 sale. Processors often keep that fee when you refund; Stripe's refund documentation, for example, says its original processing fees are not returned. On 125 orders the fees come to about $143 against $114 for 100, which cuts the lead to about $13. Add support time at, say, 10 minutes per return valued at $25 an hour (both assumptions): 11.5 extra returns take nearly two hours, about $48, and the glossy images are now about $35 behind. Then add the thing the table cannot show: a chargeback or a run of one-star reviews. Our worksheet on dropshipping profit margin after ads, fees and refunds shows how to build these allowances into a per-order margin so you can test your own numbers.

Labelling AI-made images for sales channels and buyers

Rules about labelling AI imagery are changing quickly, and they differ by sales channel and by place. Three practical layers apply to most stores:

1. Channel rules. Google Merchant Center requires images made with generative AI to carry metadata identifying them as such (its guidance names the IPTC DigitalSourceType value "TrainedAlgorithmicMedia") and says not to strip those tags, including from images made with its own Product Studio. The same page lists a "CompositeSynthetic" value for an image that combines real and synthetic elements, which describes a real product photo on an AI-made background more closely. Marketplaces and ad platforms each have their own rules. Read the image policy of every channel you sell on.
2. Advertising and consumer-protection rules. Whatever the tool, a picture that misleads buyers about the product can be treated as a misleading advert. Your adviser can tell you what applies where you sell.
3. Your own honesty standard. A short line under a lifestyle image, such as "Scene is illustrative; product shown is the real item", is cheap and helps both trust and complaints.

Keep metadata intact. Many editing and compression tools strip it by default, so after you export, check a sample image with a metadata viewer to see whether the tags survived. If your workflow strips them, adjust the export settings or add the tag back before uploading to a channel that requires it.

Alt text and filenames that match the true product

An image is not finished until it can be described. Shopify's alt-text field takes up to 512 characters, with about 125 recommended. Write what the picture shows: "Navy canvas backpack, front view, with two side pockets and a black fastener". Do not stuff keywords, and do not describe features the product lacks. An assistant can draft alt text from the real photo, but read it against the item: AI descriptions confidently add details such as "leather trim" that the product does not have.

Write alt text (under 125 characters) for each of these 5 product photos.
Describe only what is visible. Do not infer materials, brands or features that cannot be seen.
Use plain wording. Product: [name]. Return a numbered list.

Check the rewritten text against the photos, and keep filenames plain and truthful too: "navy-canvas-backpack-front.jpg" beats "best-cheap-backpack-2026.jpg".

A first-week plan for one product

Here is a modest plan to try with a single product from a four-product dropshipping store that sells home organisers.

• Day 1. Email the supplier asking for written permission to use their images, and for the original high-resolution files. Order a sample.
• Day 5 (sample arrives). Unbox on camera, note any difference from the supplier's pictures (colour, size, missing parts). Write those differences into your product page notes.
• Day 6. Shoot six photos. Use AI to clean backgrounds and make two lifestyle scenes with the product locked.
• Day 7. Check every image against the sample. Log sources. Upload with alt text. Add a one-line disclosure under the lifestyle shots.

Then split-test for two weeks: half the traffic sees the supplier's permitted images, half sees yours. Compare conversion rate and, importantly, returns and messages from buyers about "not as pictured". The pictures are only one half of a product page; the words need the same care, as rewriting supplier copy with AI explains.

Three products, three different answers

The right level of AI help depends on the item, so here are three quick walk-throughs, all illustrative.

A plain stainless-steel water bottle. Few details to get wrong: one shape, one finish, a logo-free body. Shoot the sample against a pale sheet, let AI lift the background and add a gym-bag scene. Risk is low; the one thing to check is the finish, because brushed steel and polished steel photograph very differently. Compare the shine in your final image with the bottle in your hand.

A phone case with a printed pattern. The print is the product, and a model that redraws it, even slightly, changes what the buyer is paying for. Keep every pixel of the case itself from your own photograph, and generate nothing inside its outline. Fit is also a returns driver, so state the exact phone models in the title and the first image, not only in the description.

A clip-on desk lamp with a USB cable. Ports, cable length, clamp opening and brightness are all features a buyer will check against the picture. Use real photos for the gallery, put a coin or a ruler beside the clamp for scale and treat any AI lifestyle scene as decoration. Do not let a generated scene show a wireless charging pad, a second cable or a power adaptor that the box does not contain.

The pattern across all three is the same: the more a buyer can verify with a ruler or a socket, the more the photograph must be real.

A ten-minute phone set-up that beats most supplier photos

Many owners hesitate because they imagine a studio. You need much less. Lay a sheet of plain white card on a table next to a window and prop another sheet vertically behind the product so there is no visible horizon. Put the phone on a stack of books, with the lens level with the product, and turn off the flash. Take each shot twice, one with the product centred and one with a little space around it for cropping. Use the phone's grid lines to keep the product level.

Take a close-up of every label, seam and port, because those are the pictures that reassure buyers and the ones an AI model is most likely to get wrong. Then feed the best three shots to your editing tool. The whole session takes about ten minutes of shooting plus the time to edit, and the result is something no competitor selling the same item can copy.

Where it went wrong: the sofa cover that was never that colour

An illustration, not a case. A store selling stretch sofa covers generated lifestyle images from text prompts because the supplier's photos were poor. The images showed a deep charcoal cover with a textured weave; the real product was a flat, slightly shiny dark grey. Orders arrived, buyers complained and the return rate on that product reached nearly one in five. The owner's first response was to tweak ads; the real fix was to order the sample, photograph it honestly and use AI only to clean the background.

The lesson generalises. If the first sentence of a complaint is "not as pictured", the image is the defect. Track a "not as described" reason code on every return, so you can see which product pages deserve a re-shoot before you spend another dollar on ads.

When a real photo is not possible: a limited-use rule

Sometimes you cannot get a sample. In that case, be explicit about limits. Use the supplier's licensed photos for accuracy, add AI-generated scenes only for context (a room, a desk), keep the product itself untouched, and state on the page that the scene is illustrative. Avoid text-only product generation for anything where detail matters: clothing fit and fabric, electronics ports and buttons, items with printed text, children's products. For low-detail items such as a plain mug, a generated scene around a licensed photo is a lower risk.

For fashion in particular, AI model shots raise further questions about fit and sizing; the tutorial on AI model shots for boutiques covers what try-on images can and cannot promise. And whichever approach you take, read online shop AI mistakes that hurt trust and conversions before you publish a full catalogue of generated pictures.

If you want help choosing which products justify a real shoot and which can stay on licensed images, that is a typical question for a 1:1 AI implementation call, though the workflow above is enough to start on your own.

Checklist before an image goes live

• Written permission or my own photograph behind every starting image.
• I have held a sample and compared the final image with it, at 200% zoom.
• Colour, size, shape, features and included items match the real product.
• Props are not implied to be part of the offer.
• AI-generated or edited images keep any required metadata on every channel that asks for it.
• Alt text describes only what is visible; filenames are plain.
• The source and tool for each image are logged.
• Returns are tagged "not as described" so I can learn from them.

Questions about AI product photos for dropshipping

Can I use the supplier's photos without asking?

Do not assume so. Photos on a supplier's listing or a marketplace page can belong to a photographer, the factory or a third party, and a supplier's catalogue does not always carry the right to pass them on. Get permission in writing, covering your use and your sales channels, or take your own photos.

Does Google Merchant Center accept AI-generated product images?

Its image guidance says images made with generative AI must keep metadata marking them as AI-generated, such as the IPTC DigitalSourceType value TrainedAlgorithmicMedia, and that tags should not be stripped. The main image must also show the exact item being sold. Read the current help page before uploading.

Is it enough to put 'AI generated' in small print?

Disclosure helps but does not cure a misleading image. If the image shows a colour, size or feature the product does not have, a footnote will not stop returns or complaints. Fix accuracy first, then disclose where the platform or your own policy calls for it.

What is the cheapest safe way to start?

Order one sample of each product you plan to sell, photograph it on a plain background with a phone in daylight, and use AI only to clean the background and make lifestyle scenes around your real photo. A sample costs a few dollars and removes most of the accuracy risk.

Will AI images lower my conversion rate?

It depends on how real they look and whether buyers feel misled. I cannot give a figure without your own data. Run a split test of real photos against AI-assisted ones on the same product and compare conversion and returns, not only clicks.

Further reads

• Google Ads for Dropshipping With AI: What to Let It Write — Decide what AI writes for adverts and what you check.
• Dropshipping Returns and Refunds: Policy Wording and AI Replies — Handle the returns that inaccurate images cause.
• Dropshipping Customer Service With AI: Can It Handle Slow Delivery? — Answer buyers who say the item is not as pictured.
• AI Dropshipping Product Finder: Check Demand and Supplier Risk First — Check supplier risk before you list a product.
• Taking a Local Shop Online With AI: Product Listings in a Weekend — Put a first catalogue online quickly with AI.
• Shopify Dropshipping Automation: What AI Runs and What Breaks — See what AI can run in a Shopify dropshipping store.
• Free AI Product Photoshoot: What Free Plans Really Allow — What free AI product-photo plans allow, where the limits bite, how to test them in 30 minutes and when a paid plan costs less than your time.
• Google Merchant Center Image Requirements for AI-Generated Product Photos — A pre-upload checklist for Merchant Center product images: size and format limits, the banned overlays, and how to test that AI metadata survived.
• Google Merchant Center Suspended? Fix Misrepresentation With AI Help — A step-by-step audit of the pages, details and data Google checks after a misrepresentation suspension, plus appeal wording and a scoring sheet.
• Remove Backgrounds From Product Images With AI: Glass, Hair, Shadows — One-click background removers fail on glass, fur and shadows. Here is how to shoot, remove, repair and check edges, then batch the rest.
• AI for a Dropshipping Business: Where It Helps and Where It Fails — Where AI helps an existing dropshipping store, from product copy to supplier emails, and where it cannot fix reliability or margins.
• Does Amazon Allow AI-Generated Product Images? What Passes Review — Amazon's image guide does not ban AI images, but it does ban inaccurate ones. A rejection checklist built around one listing.
• Does TikTok Shop Allow AI Content? A Label Checklist by Video Type — A label-or-not checklist for each kind of TikTok Shop video, based on TikTok's AI rules and press reports of its seller rules.
• How Does AI Dropshipping Work? One Store's AI Jobs and Approval Steps — One dropshipping store's jobs sorted into AI-does and owner-approves, with the approval step for supplier vetting, listings, support and ads.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Merchant Center help on product image requirements and AI-generated image metadata (read November 2026); Shopify help on its file editor and image generation; rights and disclosure rules vary by sales channel and place, so read each channel's current policy and ask a qualified adviser about licences.