Alibaba reverse image search: a supplier CSV with price and MOQ

Disclosure: I build TubeText Labs' Alibaba & Yiwu Supplier Finder by Image. This is a documentation-based workflow, checked on October 23, 2026, not a new sourcing test or customer case study. The inputs and decision example below are illustrative.

You have a product photo and need a shortlist of wholesale listings, with minimum order quantity (MOQ), price and a link back to each supplier. Searching only by an English product name can miss the visual details you care about.

This walkthrough uses a product image to discover similar listings on Alibaba.com and Yiwugo, then turns the results into a reviewable CSV. It stops at candidate discovery: the matches do not establish who made the photographed item, whether the goods are identical, or which supplier you should pay.

Which Alibaba image search do you need?

There are two different jobs behind this search phrase:

• Search an image collection you manage. Alibaba Cloud's Image Search documentation describes searching within a specified Image Search instance.
• Discover supplier listings from a product photo. The TubeText Actor searches Alibaba.com and Yiwugo and returns listing and supplier fields. It is an independent Apify tool, not Alibaba's official API.

This article covers the second job. It does not search 1688, Taobao, JD or AliExpress. An AliExpress product can supply a reference image; that does not make AliExpress one of the supplier-search destinations.

1. Start with one public product image

In the Actor's Input tab, use Product image URLs, not the product page URL. Choose a clear photo of one item that you have permission to use and share. Avoid private customer photos or confidential designs.

Illustrative input, not ready to run: replace the placeholder with your permitted, publicly accessible image URL.

{
  "imageUrls": ["https://example.com/replace-with-your-product-photo.jpg"],
  "sources": ["alibaba", "yiwugo"],
  "maxResultsPerImage": 5,
  "language": "en"
}

Here, maxResultsPerImage limits matches per image and per source. Five is a small starting limit, not a promise that either source will return five usable candidates. English mode adds Yiwugo's English titles and shop names while retaining the Chinese originals; Alibaba.com results are documented as English.

The input supports separate price filters for Alibaba (minPriceUSD, maxPriceUSD) and Yiwugo (minPriceCNY, maxPriceCNY). For a first pass, leaving them unset can help you see whether the photo returns the right product category before narrowing on listed price.

When you choose to run it, review the current event pricing first. Each selected source is a separate image search. Match limits, extra-match events and Actor Start charges matter; fewer returned rows do not automatically mean the entire run is free.

2. Build a CSV that preserves the evidence

After a run finishes, inspect the dataset before exporting CSV or Excel. Keep JSON as well if you need array fields such as certifications.

The documented output fields support this shortlist:

Review question	Fields to keep	How to use them
Which image produced this candidate?	input, sourceImage, scrapedAt	Preserve the reference and observation time.
Where is the listing?	source, productId, url, urlEn	Use source plus listing identity, not the title alone. urlEn is for Yiwugo.
Is it visually relevant?	title, image, rank, status, matchConfidence, similarity	Open the listing and compare the item, not just its rank. similarity is Alibaba-specific.
What price is being advertised?	price, currency, priceMin, priceMax, priceText, priceCNY	Keep currency and any range or original text. Missing is not zero.
Does the order size fit?	moq, moqUnit, moqText	Check the unit and original wording together.
Who is behind the listing?	shopName, shopUrl, supplierCountry, supplierProfileUrl	Preserve the source links for later review. Some fields are source-specific.

All of these fields are optional in the public schema. Do not assume a complete row. Keep non-ok statuses and errors in a separate review tab instead of silently deleting them or treating them as a zero-priced offer.

Add your own columns for requested_quantity, variant_checked, unit_checked, review_decision and review_notes. These are reviewer-added columns, not Actor output. If you collect several photos, keep each image-to-listing relationship even when the same listing appears more than once.

3. Filter for MOQ before ranking by price

Suppose your intended order is 200 pieces. This is a made-up decision example, not returned data:

• Candidate A has an MOQ of 1,000 pieces. A low displayed unit price does not make it suitable for the 200-piece brief.
• Candidate B has an MOQ of 100, but its unit is boxes. Its quantity is not comparable until you know pieces per box.
• Candidate C shows no price. Keep it as “quote needed,” not “free.”

Use the same checks on actual rows:

1. Open the listing and confirm that the photo match is the intended item, rather than a visually similar accessory.
2. Check variant, size, material and pack size. A shared image does not prove equivalent goods.
3. Compare MOQ only after resolving the unit. Preserve the original moqText when parsing is uncertain.
4. Compare advertised prices within the same currency and price basis. A low end of a range may not apply to your quantity or variant.
5. Record which candidates need clarification before they enter a price comparison.

Alibaba USD prices and Yiwugo CNY prices should not be sorted together as raw numbers. Even after currency conversion, a listing price is not a landed quote: shipping, packaging, duties, taxes, samples and payment terms still need separate confirmation. Subtracting it from a retail price does not establish profit.

4. Keep supplier verification separate

Alibaba-specific fields can include supplierYears, supplierRating, supplierReviews, goldSupplier, certifications and isAd. Yiwugo can include a market booth or location.

Use those fields to decide what to inspect, not as proof of manufacturing ownership, product quality or regulatory compliance. A certification label in a listing is not a verified certificate for your exact product and destination. A high visual score is not a supplier trust score.

The next stage is a reviewed specification, a supplier quote and, when appropriate, sample and document checks. This workflow does not contact suppliers, order samples or make a purchasing decision for you.

5. Already have an AliExpress product dataset?

The existing products-to-suppliers Python recipe first collects AliExpress product candidates, then passes that run's dataset to this image-search Actor. You do not need that first step if you already have a suitable image.

For an existing dataset, the supplier Actor accepts datasetId and maxDatasetItems. The schema describes using each product's main image and carrying its title, URL and price into the results. Check those images and cap the number of products before scaling. Dataset access may require the read-only permission shown by Apify.

The recipe's console output is deliberately minimal. Use the full supplier dataset for review, retaining source links, statuses, units and currencies as above. Its two Actor calls have separate event charges. The recipe was inspected for this article but not executed, and it has not been turned into a verified purchasing workflow.

Next step: open the Actor input, review the pricing, and start with one permitted product photo when you are ready. The useful first result is a small, traceable candidate list you can review, not a claim to have found the original factory.