Amazon Listing Images From a Link

Paste an Amazon URL or a bare ASIN and Graflio reads the live listing, its reviews, the category best-seller list and ten competing listings, then writes an image brief from what it found and renders the set. The whole research pass costs 0.1 credit and takes fourteen steps. This page walks the chain end to end, names what is read and what is not, and says plainly where it fails.

Why "from a link" is a real distinction

Almost every tool in this category starts from a photograph. The photograph tells the model what the product looks like. It does not tell the model what buyers complain about, what the competitor two ranks above you does better, or which of your six images should carry which argument. Those answers sit publicly on the listing page and the ten pages around it, and a link is the cheapest way to hand them over.

So the useful question is not "does it accept a URL" but "what does it do with what it reads". A tool that pulls your title and bullets to seed a prompt is doing something much smaller than one that cross-reads a hundred reviews across eleven listings before deciding what the images should say.

The chain, step by step

1. The link is normalised down to the canonical /dp/<ASIN> form — the long ref= and pd_rd_* URLs are themselves a bot-check trigger. The marketplace comes from the domain, so an amazon.de link is fetched with German headers and comes back in German.
2. Your listing is read. Title, brand, price, star rating, ratings count, the feature bullets, the product-overview spec table, up to twelve gallery images at their highest available resolution, the colour variations with their own ASINs and swatch photos, and the customer reviews Amazon renders on the page itself.
3. The category is identified from the Best Sellers Rank links, which name the exact deep subcategory. When those are missing, the breadcrumb browse-node ids are the fallback. The deepest category wins, because "Dog Chew Toys" is worth more than "Pet Supplies".
4. The best-seller list is fetched for that category. Page one renders thirty ranked items server-side. If your own ASIN is in the list, your rank is recorded.
5. Ten competitors are read in full — the same parse as your own listing, one at a time, individual failures skipped rather than aborting the run. Ten is a deliberate ceiling: twenty produced richer pain points but left the model so little room that it regularly ran out mid-answer.
6. The brief is written. One long-context model reads all eleven listings at once and returns four to seven pain points, each tagged as coming from your reviews, a competitor's, or both, and each backed by its evidence with the competitor named by brand. Then four to six USPs grounded in the data, a one-sentence positioning angle, a colour and type system derived from the product's own look, and six or seven image scripts — headline copy, two to four callouts, composition and a background treatment each.
7. The set renders. Your real product photograph is attached to every image call with the instruction to reproduce the product exactly; generation paints the scene around it. You pick one of four style modes, and any single image can be re-rendered on its own without touching the rest.

The instruction that matters most in step six is a prohibition: every statistic, percentage, review count, price and guarantee in the brief and the image scripts must appear verbatim in the scraped data. That is what stops a generated infographic claiming a 30-day guarantee you never offered.

What is read, and what is not

Roughly a hundred reviews across eleven listings is the working sample: far less than a review-mining tool sees, far more than a designer reads before opening Photoshop.

Where it fails

Amazon blocks the fetch. Amazon rate-limits datacentre traffic aggressively, so requests run through a residential proxy with rotating browser fingerprints and four attempts on progressively backed-off delays. When it still fails you are told Amazon is rate-limiting and to retry in a few minutes, rather than being handed a broken half-listing. A research job that fails refunds its cost automatically, in full, once.

The product has no reviews and no rank. A pre-launch ASIN gives the research almost nothing of its own. The competitor half still works and is often the more valuable half, but what you get is a brief built on the category, not on your product.

The page parses oddly. Amazon ships several markup variants. When the parse returns no title or images, the run retries against a different URL form and fingerprint, and failing that you enter the product details by hand.

Who else starts from an ASIN

We are not the only tool that accepts one, and any page claiming otherwise is wrong. Ecomtent's published workflow is import an ASIN, generate, refine, publish to Seller Central — $599 a month for the Seller plan and 25 SKUs, as of August 2026. Amazon's own AI Creative Studio starts from a product in your advertising catalogue and is free with an Ads account.

Which link to paste

Paste the listing in the marketplace you want images for — the brief follows the language of the listing, so an amazon.es URL produces Spanish copy with no translation step. Selling the same product across several European marketplaces, research the one with the most reviews and translate the finished images afterwards. If your ASIN is a variation child, paste the specific child you sell: its reviews and its swatch are what the images get built from. If the product is not live yet, paste your closest competitor and read the brief as a map of the category rather than a verdict on you.

Common questions

Do I need Seller Central access to use a link?

No. Everything read here is on the public product page, which means you can research a competitor's ASIN exactly as easily as your own. It also means nothing from your Business Reports — sessions, conversion rate, returns — is available to the research.

How many reviews does it actually read?

Around eight to ten per listing, which is what Amazon renders on the product page. Across your listing plus ten competitors that is roughly a hundred reviews. The dedicated review pages require a login, so no tool reading a public URL is reading thousands.

Does it work on amazon.de, .fr, .es and .co.uk?

Yes. The marketplace is taken from the domain and the fetch uses that marketplace's language, so the listing, the best-seller list and the resulting brief all come back in the local language.

What does the research pass cost?

0.1 credit per product, and a failed job is refunded automatically. Every account starts with 30 credits, so the research itself is effectively free until you start rendering images.

Can it generate images without the research?

No. The image scripts are the output of the research step, so the set is always built from a brief rather than from a prompt you type. If you want to render from a prompt of your own, this is the wrong tool.