Google Images Scraper — Image Search with Full Metadata

Get what Google Images shows for any search term, one row per image: the full-size image URL (Google's link to the original file, not its thumbnail) with its width and height, file size, format and dominant color, the page it appears on with its title, site name and domain, and — when Google has them — the product on the page (title, brand, price, currency) and the image's license details (license link and Creative Commons license name, creators, credit line, copyright notice). Links that only crawlers can open (some Instagram, Facebook and TikTok images) are flagged, and can be left out.

Google's own filters (size, color, type, file type, time, usage rights), in 51 countries and the interface language you choose. When Google runs out of images for a term before your limit, the Actor can continue with the related searches Google suggests for it ("labrador golden retriever" for "golden retriever"), minus the ones you exclude; every row says which search found it.

No Google account, no API key, no browser. Requests go through Apify's Google SERP proxy.

Quick start (API)

curl -X POST "https://api.apify.com/v2/acts/foxlabs~google-images-scraper/run-sync-get-dataset-items?token=YOUR_APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"queries": ["golden retriever"], "maxImagesPerQuery": 50}'

What you get

Group	Fields
Image	imageUrl (Google's link to the original file), imageUrlCrawlerOnly (true for Instagram, Facebook and TikTok crawler links that return a web page or an error instead of the file — see below), imageWidth, imageHeight, aspectRatio (width ÷ height), orientation (landscape, portrait, or square when width and height differ by less than 5%), format (jpg, png, gif, webp, svg… from the file name; null when the URL has no image extension, e.g. .php or none), fileSizeBytes (from Google's label: 1 KB = 1,024 bytes, rounded by Google), fileSizeText ("522KB"), dominantColor ("rgb(89,124,121)") and dominantColorHex ("#597c79"), thumbnailUrl, thumbnailWidth, thumbnailHeight
Source	title, pageUrl (the page the image appears on), domain (that page's site), siteName, aboutThisImageUrl (Google's "About this image" page)
Product	isProduct, productTitle, productBrand (the brand Google lists: "Nike" for Nike shoes on Foot Locker or eBay; for marketplace sellers such as Etsy shops it is often the shop's name), productPrice, productCurrency (as Google shows it for that listing — it can differ from your country), productPriceText ("$140.00"), productDescription, productId, productGoogleShoppingUrl
License	isLicensable, licenseUrl (the license or licensing page the site gives), licenseName ("CC BY-NC 2.0", "CC0 1.0", "Public Domain Mark 1.0" — only for creativecommons.org licenses), licenseCreators[], licenseCreatorsText, licenseCredit, licenseCopyright
Context	type (image, status or relatedSearches), query (your term), sourceQuery (the Google search that returned the image: your term or a related search), fromRelatedSearch, position (Google's order within sourceQuery), searchPage, imageId (Google's image ID, unique per term), country, language, filters, searchUrl (the Google page it came from), scrapedAt, error (status rows only), relatedSearches (related-searches rows only)

Product and license fields are filled only when Google shows that information for the image; otherwise they are null, isProduct and isLicensable are false, and licenseCreators is []. The file extension usually matches the real file type, but not always (2 of 34 files we downloaded).

Sample output

Real rows from two test runs of build 0.1.3 on 2026-09-30 (UTC), runs Wm3JXc2AkADiotOj6 and ATqAHYMQiz25AIUKa. The first row is complete except for its eight null product fields; the second row shows only its image, page and product fields.

{
  "type": "image",
  "query": "golden retriever",
  "sourceQuery": "golden retriever",
  "fromRelatedSearch": false,
  "position": 86,
  "searchPage": 2,
  "title": "Cuddly Golden Retriever Puppies Are the Cutest Thing You'll See Today - A-Z Animals",
  "imageUrl": "https://a-z-animals.com/media/2025/10/shutterstock-2300276251-medium-licensed.jpg",
  "imageUrlCrawlerOnly": false,
  "imageWidth": 1000,
  "imageHeight": 667,
  "aspectRatio": 1.499,
  "orientation": "landscape",
  "format": "jpg",
  "fileSizeBytes": 44032,
  "fileSizeText": "43KB",
  "dominantColor": "rgb(88,85,75)",
  "dominantColorHex": "#58554b",
  "thumbnailUrl": "https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcQbL3vH8Mb-yO3WPRBUBYdeG35wEN8_5_rX3RZkrKAdFg&s=10",
  "thumbnailWidth": 678,
  "thumbnailHeight": 452,
  "pageUrl": "https://a-z-animals.com/articles/cuddly-golden-retriever-puppies-are-the-cutest-thing-youll-see-today/",
  "domain": "a-z-animals.com",
  "siteName": "A-Z Animals",
  "aboutThisImageUrl": "https://www.google.com/search/about-this-image?img=H4sIAAAAAAAA_wEXAOj_ChUI2JrLx_mxvJkZEN_PpaWpqbuDsQH-W0u0FwAAAA%3D%3D&q=https://a-z-animals.com/articles/cuddly-golden-retriever-puppies-are-the-cutest-thing-youll-see-today/&ctx=iv&hl=en-US",
  "isProduct": false,
  "isLicensable": true,
  "licenseUrl": "https://www.shutterstock.com/license?utm_source=iptc&utm_medium=googleimages&utm_campaign=webstatement",
  "licenseName": null,
  "licenseCreators": ["ConstanzaMartinez"],
  "licenseCreatorsText": "ConstanzaMartinez",
  "licenseCredit": "Shutterstock",
  "licenseCopyright": "Copyright (c) 2023 ConstanzaMartinez/Shutterstock. No use without permission.",
  "imageId": "32eplErtBrE7rM",
  "country": "us",
  "language": "en",
  "filters": null,
  "searchUrl": "https://www.google.com/search?q=golden+retriever&udm=2&hl=en&gl=us&start=10",
  "scrapedAt": "2026-09-30T21:55:37.226Z"
}

A product image (search "nike air max 90"):

{
  "type": "image",
  "query": "nike air max 90",
  "position": 8,
  "title": "Nike Air Max 90 (Green Bay Packers) Men's NFL Shoes - Vintage Green/University Gold/Black/Alabaster",
  "imageUrl": "https://static.nike.com/a/images/t_web_pdp_936_v2/f_auto,u_9ddf04c7-2a9a-4d76-add1-d15af8f0263d,c_scale,fl_relative,w_1.0,h_1.0,fl_layer_apply/357105d2-f234-4916-807a-b9e312aa5bb1/NFL+RIVALRY+AIR+MAX+90+PACKERS.png",
  "imageUrlCrawlerOnly": false,
  "imageWidth": 1872,
  "imageHeight": 2340,
  "format": "png",
  "fileSizeText": "266KB",
  "pageUrl": "https://www.nike.com/t/air-max-90-green-bay-packers-mens-nfl-shoes-YXEcR3is",
  "domain": "www.nike.com",
  "siteName": "Nike",
  "isProduct": true,
  "productTitle": "Nike Air Max 90 (Green Bay Packers) Men's NFL Shoes - Vintage Green/University Gold/Black/Alabaster",
  "productBrand": "Nike",
  "productPrice": 140,
  "productCurrency": "USD",
  "productPriceText": "$140.00",
  "productDescription": "The Air Max 90 stays true to its running roots with the iconic Waffle sole, while stitched overlays and textured accents create the '90s look you ...",
  "productId": "4371935337200118775",
  "productGoogleShoppingUrl": "https://www.google.com/search?q=nike+air+max+90&prds=headlineOfferDocid:4371935337200118775,productDocid:4371935337200118775,pvo:4&ibp=oshop"
}

How it works

• Google's own pages first. A Google Images results page has up to 100 images (sometimes 50). The Actor reads the term's pages (start = 0, 10, 20…) until it has enough images or Google has no more; after the first page, it reads two pages at a time while more than 100 images are still missing. How many images Google has for one term varies: "golden retriever" stopped after 237 new images (3 pages), "eiffel tower" after 4 pages, while "mountain landscape" still had more after 400 (5 pages) — runs VK6sfzhgYpfHzGM46 and WXhwgCYvZmnLIecCD. Later pages often repeat earlier images. If one of the later pages fails twice, the term's own pages end there and the SOURCE_REPORT note says so.
• Related searches (relatedSearches: auto, the default). Only when the term's own pages run out before your limit, the Actor continues with the related searches Google lists for the term, in Google's order, and with the related searches listed on those pages. It uses only those that contain every word of your term (words of 2+ letters, as text: "golden retrievers" counts), such as "golden retriever puppy" or "labrador golden retriever". Rows found this way have fromRelatedSearch: true and the related search in sourceQuery. Set relatedSearches: off to get only the term's own results.
• A related search can change the subject. It can contain every word of your term and still be about something else: for "red panda", Google suggested "red panda drawings", "anime red panda" and "turning red red panda" (a film). In two 500-image runs for large photos of red pandas, those searches brought 141 and 168 of the 500 rows (runs DvuCf67NZp1mpXNZ0, IlmkXb9LKfPQIaUO6). Put such words in relatedSearchesExclude (see the first example below), or drop rows by sourceQuery.
• The data comes from Google's page, not from the images. The original URL, size, file size, color, title, site, product and license details are what Google publishes with each result. The Actor does not download the images.
• No duplicates within a term. Images are de-duplicated by Google's image ID. With removeDuplicatesAcrossQueries, an image is delivered once per run, under the first term that found it.
• Only delivered images are charged. A term with no result, or where the Actor-side filters removed everything, leaves one free status row with an error explaining why. For example, minWidth: 20000 gave one row saying that Google showed 500 images and the width filter removed all 500; the Actor stopped after 3 related searches that brought nothing (6 Google searches in all, run GyVVe4hIm5khebPzp). The SOURCE_REPORT record in the key-value store lists, per term, every Google search made, the duplicates, the images left out by each filter, and the related searches Google showed, used and skipped.

Input & filters

Input	What it does	Default
queries	Search terms, one per line	—
maxImagesPerQuery	Images per term, up to 1,000. The Console form starts at 20	100
relatedSearches	auto (continue with Google's related searches when the term runs out) or off	auto
relatedSearchesExclude	Words or phrases: a related search that contains one of them is not used ("drawing", "anime", "turning red"). The skipped ones are listed in SOURCE_REPORT	—
country	Which Google to search: 51 countries, each on its own Google domain	us
language	Interface language (en, de, pt-BR…); empty = the Actor's default for the country (for example de for Germany, fr for Belgium, en for India)	—
imageSize	Google filter: large, medium, icon	any
color	Google filter: blackAndWhite, transparent, red, orange, yellow, green, teal, blue, purple, pink, white, gray, black, brown	any
imageType	Google filter: photo, clipart, lineDrawing, animated	any
fileType	Google filter: jpg, png, gif, bmp, svg, webp, ico	any
time	Google filter: pastDay, pastWeek, pastMonth, pastYear (when Google found the image; the output has no date because Google shows none per image)	any
usageRights	Google filter: creativeCommons, commercial	any
aspectRatio	Computed by the Actor from the image size: tall < 0.8 ≤ square < 1.25 ≤ wide < 2 ≤ panoramic. Google ignores its own aspect-ratio parameter, so the Actor filters by the size instead	any
minWidth / minHeight	Actor-side: keep images at least this large (pixels of the original)	0
skipCrawlerOnlyImages	Actor-side: leave out images whose imageUrl is an Instagram, Facebook or TikTok crawler link (imageUrlCrawlerOnly: true), and fill the limit with other images. Useful for image datasets	off
removeDuplicatesAcrossQueries	Deliver an image once per run, under the first term that found it	off
outputRelatedSearches	Add one free row per term (type: relatedSearches) with Google's related searches	off
maxSearchesPerQuery	Upper bound on Google result pages read for one term	30

How the Google filters were checked. Every value was tried against Google on 2026-09-30 (40 test searches), most of them next to the same search without the filter. Examples: imageSize: large → 94% of images at least 1,000 px wide (57% without); color: blue → blue-dominant images 77% (11%); fileType: svg → 93% .svg files. Weaker or less direct evidence: teal and brown raised the share of that color only to 19% (1% without); photo replaced about half of the images; animated (95% GIF files) was checked on one term without an unfiltered comparison; the time filters change the result set (1–32% of images in common with the unfiltered search), but Google shows no date per image, so their accuracy is not measured. Google's own il:cl usage-rights value and its aspect-ratio value had no effect, so they are not used. "Face" images and SafeSearch are not offered. With usageRights: creativeCommons, all 150 images in two runs carried a creativecommons.org link, 149 of them to a specific license (runs lCm3dwGyOJRkz8lyg, lfV1bcqMk7oJ2fwe9), from CC0 and public-domain marks to CC BY, BY-SA and BY-NC licenses — check licenseName before reusing an image commercially. In a 50-image Creative Commons run with build 0.1.2 (gfJQMjhMrnGl9oAbe), all 50 rows had a licenseName (22 "CC0 1.0", 19 "Public Domain", 5 "Public Domain Mark 1.0", 4 CC BY / BY-NC variants).

Invalid input never reaches Google. A value outside the lists or limits (an unknown color, maxImagesPerQuery above 1,000) is refused by Apify before the run starts, with the reason in the error. Input the Actor cannot use (for example only blank search terms) stops the run at once, with the reason in the run's status message (run IxLhC1z3qXPPDdxgZ).

Example inputs (copy & paste)

A training set of large photos, at least 1,200 px wide, without crawler-only links and without related searches that change the subject:

{ "queries": ["red panda"], "maxImagesPerQuery": 500, "imageSize": "large", "imageType": "photo", "minWidth": 1200, "skipCrawlerOnlyImages": true, "relatedSearchesExclude": ["drawing", "anime", "cartoon", "turning red"] }

In run swiuLsZv72BFrZOUO (338 s, 13 Google searches) this gave 500 unique images, all at least 1,200 px wide and none crawler-only; 107 crawler-only and 115 narrower images were left out. The list skipped 12 related searches ("turning red red panda", "red panda drawings", "anime red panda", "red panda cartoon"…) and the Actor used 8 others ("cute red panda", "baby red panda", "red panda angry"…). Rows whose title or page URL mentions drawings, cartoons, anime or the film: 22 of 500, against 123 and 131 of 500 in the two runs without the list (DvuCf67NZp1mpXNZ0, IlmkXb9LKfPQIaUO6). That is a word check on titles and URLs, not a look at the images.

Images you can reuse under Creative Commons licenses:

{ "queries": ["mountain landscape"], "usageRights": "creativeCommons", "maxImagesPerQuery": 100 }

Product images; in our 200-image test, 25% of rows had a brand and 9% a price (run oKwsyiLr5cBDKoMLM):

{ "queries": ["nike air max 90"], "maxImagesPerQuery": 200 }

Image search results in Germany for several terms, plus Google's related searches for keyword research:

{ "queries": ["hundefutter", "katzenfutter"], "country": "de", "maxImagesPerQuery": 100, "relatedSearches": "off", "outputRelatedSearches": true }

Use cases

• Image datasets for machine learning: many unique original images per term, with size, format and file size, filterable by size, type and color; skipCrawlerOnlyImages keeps only links that are not crawler-only, and relatedSearchesExclude keeps related searches on your subject.
• Image SEO and brand monitoring: which pages and domains rank in Google Images for your terms, by country, and at which position.
• E-commerce research: which sites show a product's images, and the product, brand and price Google attaches to them.
• Finding reusable images: Creative Commons results with the license name, creators, credit line and license link Google shows.
• Keyword research: Google's related searches for each term.

Performance & coverage

Measured on the Apify platform on 2026-09-30/10-01 (default 256 MB memory; build 0.1.1 unless noted):

Search	Images	Google searches	Time	Run
golden retriever, 20 images (the form's starting input)	20	1	17 s	sfIdiHLmJJa663fxw
golden retriever, 50 images (Quick start above)	50	1	11 s	862qmvHQNyFKYUc6j
golden retriever, 700 images, related searches auto	700 (237 from the term, 463 from 6 related searches)	9	221 s	VK6sfzhgYpfHzGM46
nike air max 90, 200 images	200	3	150 s	oKwsyiLr5cBDKoMLM
kedi, Turkey, 200 images	200	3	190 s	auMmWTF4QNZbtjhJP
hundefutter + katzenfutter, Germany, 100 each	200	3	52 s	7uW2bhAsVaDlxVXKa
mountain landscape, Creative Commons, 100 images	100	2	16 s	lCm3dwGyOJRkz8lyg
red panda, 500 large photos ≥1,200 px wide	500 (120 narrower images left out)	9	442 s	DvuCf67NZp1mpXNZ0
3 terms × 400 images (eiffel tower, mountain landscape, nike air max 90)	1,200	18	269 s	WXhwgCYvZmnLIecCD
golden retriever, 700 images (build 0.1.2)	700 (196 from the term, 504 from 7 related searches)	12	248 s	ndguazKi6tJfEfWix
red panda, 500 large photos ≥1,200 px (build 0.1.2)	500 (126 narrower images left out)	10	326 s	IlmkXb9LKfPQIaUO6
red panda, the first example below (build 0.1.3)	500 (107 crawler-only and 115 narrower images left out; 12 related searches skipped)	13	338 s	swiuLsZv72BFrZOUO
golden retriever, 100 images, related searches off (build 0.1.3)	100 (2 crawler-only links flagged)	2	72 s	Wm3JXc2AkADiotOj6
nike air max 90, 100 images (build 0.1.3)	100	1	27 s	ATqAHYMQiz25AIUKa

Run time depends mostly on how fast Google answers through the proxy: in the runs above, the median answer per run was 3–40 seconds, and the slowest answer logged by build 0.1.3 (which logs answers and time-outs apart) was 80 seconds. Across 120 platform test runs of builds 0.1.1–0.1.3 (the table above, 4 more checks and 102 runs of the input-combination test), the Actor made 229 Google searches: 228 succeeded, 5 of them after one retry. One related search timed out twice and was skipped; that run still delivered its 500 images from other related searches. All 7 failed attempts were stopped at the Actor's 90-second limit. The term's own pages can differ between runs: the two 700-image golden retriever runs got 237 and 196 new images from them, and later pages often repeat earlier images (39 of 50 on one page). Peak memory stayed at or below 96 MB of the default 256 MB.

Data quality

Share of images with a value (platform runs VK6sfzhgYpfHzGM46, oKwsyiLr5cBDKoMLM, auMmWTF4QNZbtjhJP, WXhwgCYvZmnLIecCD):

Field	golden retriever (700)	nike air max 90 (200)	kedi, Turkey (200)	3 terms × 400 (1,200)
imageUrl, width, height, pageUrl, title, domain, siteName, fileSize, dominantColor, thumbnail	100%	100%	100%	100%
aboutThisImageUrl	100%	100%	100%	100%
format (from the file name)	82%	78%	89%	85%
product: brand / price	2% / 2%	25% / 9%	1% / 1%	7% / 3%
license link	9%	0%	4%	10%
unique images per term	700 / 700	200 / 200	200 / 200	1,200 / 1,200

• Checked against the images themselves: for 40 sampled rows from 39 sites, our QA script (not the Actor) downloaded the linked image and read its real pixel size from the file. 34 links returned an image file, and all 34 had exactly the listed width and height; the byte size was within 5% of fileSizeBytes for 29 of the 31 files with a known size. Of the other 6, 4 sites refused the download (HTTP 403) and 2 were Instagram crawler links that returned a web page. 28 of the 40 source pages opened for the script (12 refused automated visits), and all 28 showed the image or the title.
• Crawler-only links: in the test runs of builds 0.1.1–0.1.2, 369 of 4,670 image rows (7.9%) had an imageUrl that was an Instagram or Facebook crawler link (326) or a TikTok image-API link (43); from 0% to 13% per run. In our checks these returned a web page or HTTP 403 instead of the file (Meta links 11 of 11, TikTok 2 of 2). Build 0.1.3 flags them (imageUrlCrawlerOnly) and can leave them out (skipCrawlerOnlyImages); with that option, a 100-image golden retriever run left out 3 and still delivered 100 (run aO8b7bH9oVu22baW1).
• Related searches and the subject: in the 700-image golden retriever run (VK6sfzhgYpfHzGM46), 425 of the 463 rows from related searches (91.8%) had "golden retriever" in their title or page URL, and 450 (97.2%) had "golden" or "retriever"; for the term's own rows, 229 and 235 of 237. That was one term: for "red panda", related searches changed the subject (see How it works).
• Country matters: for kedi, the first 100 images in Turkey and in the US shared 20 (runs auMmWTF4QNZbtjhJP, BVx4kKoUB8W4wbb3O).
• File size is Google's label converted to bytes (1 KB = 1,024 bytes).
• Product prices are in the currency Google shows for that listing, which can differ from your country (US searches showed prices in USD, EUR, IDR, ILS, GBP, ZAR, INR and CAD).

Pricing

Pay per event: one image event per delivered image row. Status rows and related-searches rows are free, and images left out by the Actor-side filters (minWidth, minHeight, aspectRatio, skipCrawlerOnlyImages) are not delivered, so they are not charged. Current prices, including any per-run start event, are on the Pricing tab.

FAQ

Why do I get fewer images than I asked for? Google stops after a few pages for one term. With relatedSearches: auto the Actor continues with related searches that contain every word of your term (minus the ones you exclude); when those run out too, or 3 in a row bring no image that passes your filters, the term ends. SOURCE_REPORT shows each search made and the reason in a note.

Are these the original images? imageUrl is Google's link to the original file on the site that hosts it, with the width and height Google reports: of 40 sampled links, 34 returned an image file, and all 34 had exactly the listed size; 4 sites refused the download (HTTP 403) and 2 were Instagram crawler links. Rows with such crawler links have imageUrlCrawlerOnly: true; skipCrawlerOnlyImages leaves them out. thumbnailUrl is Google's small copy.

Can I use the images? Check the license on the source page. The usage-rights filter and the license fields show what sites publish; they are not legal advice.

Why is format empty for some images? It is read from the file name, and some image URLs have no image extension (for example …/image?id=123 or ….php).

How do I get only my term's own results? Set relatedSearches to off, or keep only rows with fromRelatedSearch: false.

Does the Actor download the images? No. It returns the URLs; download them yourself if you need the files.

Why does a filtered run take longer? With minWidth, minHeight, aspectRatio or skipCrawlerOnlyImages, the Actor reads more Google pages to replace the images it leaves out (a 500-image red panda run left out 120 narrower images and made 9 Google searches, run DvuCf67NZp1mpXNZ0).

Troubleshooting

• "Google Images shows no images for this search with these filters": try a broader term or fewer filters.
• "none passed the Actor-side filters": aspectRatio, minWidth, minHeight or skipCrawlerOnlyImages removed every image Google showed; the message says how many each removed.
• "Apify Google SERP proxy is not available to this run": the Google SERP proxy must be enabled on your Apify account (Settings → Proxy).

Notes, limits & legal

• Data is what Google Images publicly shows at run time; results change over time and by country.
• Google may change its page without notice; fields can then come back empty until the Actor is updated. If you see that, open an issue with the run ID.
• Images belong to their owners. The Actor returns links and the information Google publishes about them, not the image files.
• Google and Google Images are trademarks of Google LLC; this Actor is not affiliated with Google.

Support

Open an issue on the Issues tab with the run ID and the input, or write to info@foxlabs.com.tr.

Changelog

0.1.4 — 2026-10-01

• README checked claim by claim against the test runs and corrected; no code change.

0.1.3 — 2026-10-01

• New field imageUrlCrawlerOnly and new input skipCrawlerOnlyImages (Instagram, Facebook and TikTok crawler links).
• New input relatedSearchesExclude (skip related searches that change the subject).
• SOURCE_REPORT lists every Google search; latency covers answered requests, and timeouts counts requests stopped at 90 seconds.
• An impossible Actor-side filter now reads at most 4 related searches (up to 6 before). Details in CHANGELOG.md.

0.1.2 — 2026-10-01

• New field licenseName (Creative Commons license name).
• An impossible Actor-side filter stops widening after 3 related searches in a row that deliver nothing, instead of using up maxSearchesPerQuery.
• A term's own result pages are read two at a time after the first (see CHANGELOG.md for the measured effect).

0.1 — 2026-09-30

First version. See CHANGELOG.md.