There are two ways product images get you into trouble in 2026, and most sellers only worry about one of them. The first is using photos that are not yours — supplier shots, brand catalogue images, a competitor's carefully lit hero — and discovering that the marketplace already knows the image belongs to somebody else. The second is the reverse: waking up to find your own photography powering a counterfeit listing that undercuts you by 40 per cent.
Both problems come down to the same technology. Marketplaces no longer compare filenames or file bytes. They compare what an image looks like. Understanding how that works tells you exactly what you can and cannot get away with, and exactly how to fight back when your work gets lifted.
How marketplaces actually detect duplicate images
The old assumption — crop it, flip it, save it at a different quality and it becomes a "new" image — stopped being true years ago. Detection now runs in layers.
Layer one: exact file hashing
A cryptographic hash (MD5, SHA-256) of the file bytes. Fast, cheap, and catches only untouched copies. Change a single pixel and the hash changes completely. This layer catches the laziest cases: the same JPEG uploaded to two listings.
Layer two: perceptual hashing
This is the layer that catches most sellers out. A perceptual hash reduces the image to a small fingerprint of how it looks — typically by shrinking it to something like 32x32 pixels, converting to greyscale, running a frequency transform and recording which regions are brighter or darker than average. The output is a short string, often 64 bits.
The key property is that visually similar images produce similar hashes. Resize a 3000px image to 1200px, re-save it as WebP at 80 per cent quality, brighten it slightly, and the perceptual hash barely moves. Systems then measure the Hamming distance between two hashes; anything under roughly 5–10 bits of difference is treated as the same image. Common variants you will see referenced are aHash (average), dHash (difference), pHash (discrete cosine transform, the most robust of the classic set) and wHash (wavelet).
Layer three: deep embeddings
Perceptual hashes struggle when an image has been meaningfully edited — heavily cropped, mirrored, recoloured, or composited onto a new background. Convolutional and transformer-based models solve this by converting each image into a vector of hundreds of numbers describing its content, then measuring cosine similarity between vectors. A 2026 comparative study of deduplication methods found that classical hashes remain the efficiency winner for near-identical copies, while CNN embeddings are materially better on edited duplicates — which is precisely the marketplace-theft case.
Large platforms run all three. Amazon's image systems are described as analysing pixel patterns, colour histograms, edge detection and perceptual hashing together to fingerprint how an image looks. Mirroring a supplier photo does not defeat that.
Why this hurts honest sellers, not just copycats
The most common casualty is not a counterfeiter. It is a legitimate reseller or dropshipper who downloaded the manufacturer's image pack from 1688, AliExpress or a brand portal — exactly as thousands of other resellers did — and uploaded it unchanged.
What follows tends to be one of these:
- Image rejection at upload. The image is flagged as already associated with another listing and never goes live, leaving you with a listing that has no main image and therefore no traffic.
- ASIN or listing suppression. On Amazon, duplicate detection can trigger a warning that the ASIN will be suppressed — the listing exists but stops appearing in search results. Sales fall to near zero without an obvious cause.
- Merge into someone else's listing. Identical imagery is one of the signals used to decide that two listings are the same product. You can lose control of the detail page entirely.
- An IP complaint. If the rights owner is enrolled in a brand protection programme, the image search tools inside it will find your listing. Repeat complaints escalate towards account-level enforcement.
The uncomfortable truth: a supplier saying "feel free to use our images" is not a licence you can show a marketplace, and it does nothing about the fact that 300 other sellers have the same file.
What actually makes an image count as yours
Small edits do not work. What does work is producing genuinely different visual content. In rough order of effort:
| Approach | Beats perceptual hash | Beats deep embedding | Effort |
|---|---|---|---|
| Resize, recompress, rename | No | No | Minutes |
| Mirror, rotate, adjust brightness | Sometimes | No | Minutes |
| Crop tightly to a different framing | Often | No | Minutes |
| Cut out the product, place on new background | Yes | Partly | ~15 min/image |
| Reshoot the product yourself | Yes | Yes | 1–2 hrs/product |
| Original lifestyle or infographic images | Yes | Yes | Half a day |
The practical middle path for resellers: use one supplier image if the brand genuinely permits it and it is the standard catalogue shot everyone shares, then make positions two through seven entirely your own — in-hand scale shots, packaging, detail crops, a size or specification infographic, a lifestyle frame. Those are the images that both differentiate you in the gallery and carry no duplicate risk.
Whatever route you take, the output still has to meet each channel's dimension and file-size rules, and doing that by hand across five marketplaces is where most of the time goes. Batch-resizing a finished set to Shopify, Shopee, Lazada, Amazon and Carousell specifications in one pass with a tool like PixelPrep keeps the effort in the photography, not the file wrangling.
The other side: finding and removing stolen copies of your photos
Step 1 — search for them properly
Run your main hero image through three engines, not one:
- Google Lens / Google Images. The widest index, and the best first stop for anything on a publicly crawled page.
- TinEye. Specialist in exact and near-duplicate matches, with a "sort by oldest" option — the single most useful feature in a dispute, because it evidences that your version existed first.
- Yandex Images. Consistently surfaces non-Western sites that Google under-indexes, which is where a lot of counterfeit storefronts sit.
Do this quarterly for your top 10 sellers rather than trying to police the whole catalogue.
Step 2 — build the evidence before you report
Every platform wants the same things. Capture, in one folder per case:
- A screenshot of the infringing listing showing the seller name, the listing or ASIN identifier, and your image in use.
- The URL of the infringing listing.
- The URL of your original listing or website page.
- Proof of authorship — the original RAW or full-resolution file, with its capture date, and any registration or licence documents.
Step 3 — report through the right channel
| Platform | Route | Notes |
|---|---|---|
| Amazon (brand enrolled) | Brand Registry – Report a Violation | Includes proprietary text and image search across the catalogue; reports feed predictive automation |
| Amazon (not enrolled) | Public Report Infringement form | Slower, no bulk search tooling |
| Shopee | Brand IP Portal (brandipp.shopee.com) | Register the IP first, then submit via Case Management; accepts trademark, copyright, patent, industrial design |
| Lazada | IPR Protection programme complaint | Does not proactively scan listings — nothing happens until you complain; penalised sellers have products locked and revenue withheld |
| Own-domain copycats | DMCA notice to host and to Google | Also file with the payment processor where relevant |
One caveat worth knowing: a DMCA takedown can be met with a counter-notice. If the other party files one, you generally need to show you have commenced legal action or the content gets reinstated. Registering your key product photography formally with a copyright office — cheap, and done in batches — makes a counter-notice far harder to sustain.
Step 4 — make your images self-identifying
Embed IPTC metadata into your masters before they ever leave your machine. The relevant fields are Creator, Copyright Notice, Credit Line, Web Statement of Rights and Licensor URL. Two practical benefits: the metadata travels with the file when someone re-uploads it, and Google Images will display a "Licensable" badge plus a link back to your licensing page when Web Statement of Rights and Licensor URL are present. It is quiet, permanent attribution that costs nothing per image once your export preset is set up.
Your duplicate-image checklist
- Never upload an unmodified supplier or brand file as your main image on a marketplace that fingerprints images.
- Assume mirroring, cropping and recompressing will not defeat detection — reshoot or recomposite instead.
- Shoot at least four original images per SKU: in-hand scale, packaging, a detail crop, and a lifestyle or infographic frame.
- Keep full-resolution masters with intact capture dates as your proof of authorship.
- Embed IPTC Creator, Copyright Notice and Web Statement of Rights in every export.
- Reverse-search your top 10 hero images every quarter across Google Lens, TinEye and Yandex.
- Enrol in Brand Registry or the Shopee Brand IP Portal if you own a brand — the search tooling alone justifies it.
- Build one evidence folder per infringement case before you file anything.
- Register your best-performing product photography with a copyright office to survive counter-notices.
- If a listing suddenly loses all its traffic with no policy notice, check whether the main image has been flagged as a duplicate before you blame the algorithm.
Original photography has always been the strongest conversion lever available to a small seller. In 2026 it is also the cheapest form of listing insurance you can buy.