Photo Search takes up to four photos, identifies what the item actually is, and prices it against comparable listings that are live on eBay right now. This guide covers how the identification is built, where the price figures come from, and the limits worth knowing before you trust a number.
Photo Search turns up to four photos of an item into an identification, a market read from active eBay listings, and a ready-to-paste listing draft. It is built for resellers who are holding the item in their hands and want the boring part done in seconds.
Asking prices from active eBay listing data — a calculated market estimate, not completed sales.
Every run produces four things. First, an identification: brand, model or set number, edition, and any legible markings, expressed only from what the photos actually show. Second, a search query built from that identification, which is fed straight into the same eBay search pipeline the rest of the site uses. Third, price guidance — the typical asking price across comparable active listings, the range those listings cover, and a suggested listing price you can adjust. Fourth, on request, listing copy: an eBay-shaped title inside the character limit, a description, item specifics drawn from the identified attributes, and condition wording that matches what you selected.
Your saved intro paragraph and signature from Settings are appended automatically when you have them enabled, so the generated description reads like yours rather than like a template.
Valuation is drawn from the listings eBay itself returns as visually similar to your photo. Those results are put through the same filters used everywhere else on the site — accessories, spare parts, manuals, box-only and parts-or-repair listings are rejected, as are listings carrying a conflicting model number — and what survives becomes the comparable set, closest visual match first. Every price panel states its basis, so you always know whether you are looking at visually matched or keyword-matched listings.
If too few visual matches survive filtering, the search falls back to the keyword ladder built from the identification and says so. Visual and keyword comps are never blended into one unlabelled estimate. Below three usable comparables no estimate is computed at all: the listings found are shown with their real prices and conditions, labelled as an insufficient sample.
A switcher under the identification lets you move between visual match, keyword match, and an exact model or UPC lookup at any time. Switching re-prices immediately and the switcher stays put, so you can try another approach without starting over.
Photo Search detects automatically whether it is looking at one item or a pile. A single item gets a focused identification and one listing draft. A mixed lot gets an inventory read: the individually identifiable pieces are listed out, a spotted floor value is assembled from the pieces the model can actually name, and the listing copy is written to sell the group as one lot rather than pretending it is a single product.
You can override the detection either way. If the model calls a shelf of loose figures a single item, switch it to lot mode and re-run; nothing is re-photographed.
Input: four photos of a sealed boxed building set — front of box, back of box, the printed set number on the corner, and the barcode panel. The identification returns the brand, the four-digit set number, the set name printed on the box, and 'sealed, box shows shelf wear' from the visible corner crush. Because the confidence is high, the search runs automatically.
Output: typical asking price across comparable sealed listings, the low-to-high range, an active listing count, and a suggested listing price a few percent under the typical ask if you want it to move. The listing draft leads with brand, set number and name, states 'sealed, minor shelf wear to box corners' in the condition line, and fills item specifics with brand, set number, piece count where legible, and theme.
The identification asserts only what is legible. If a model number is blurred, it says the model is uncertain rather than guessing a plausible one — an uncertain attribute is left out of the search query instead of narrowing it wrongly. If a first search returns nothing usable, the query is progressively simplified until it returns comparable listings, and you can always edit the query by hand.
Price guidance is built from active listings, which are asking prices. Sellers ask for more than items sell for, so treat the numbers as a market read rather than a sale price. Comps that are clearly the wrong item can be dismissed individually, which recalculates the guidance.
Photo Search needs a free account, and daily usage is limited on the free plan.