An image scraper API turns a search query into a list of image results with their sources — title, thumbnail URL, the page hosting the image and that page's host — so you can build an image dataset without driving a browser through Google Images.
The honest framing matters here: this returns image results and their provenance, not a zip of files. Every row tells you what the image is called, where its thumbnail lives and — crucially — which page it came from. That last field is what turns a pile of pictures into a dataset you can attribute, licence-check and cite. Full-size URLs come back only when Google exposes them, because inventing them would be worse than omitting them.
$0.0003 per delivered image, up to 100 images per run. Nothing delivered means nothing charged.
A query goes to the Google Images vertical from an exit in your chosen country, and each result becomes a row. The whole run happens on the HTTP tier — no browser, no scrolling, no waiting for lazy-loaded tiles — which is why it stays at three hundredths of a cent per image and why a 100,000-image dataset is a budget line rather than a project.
Two design decisions are worth calling out. First, source_url and source are first-class fields rather than an afterthought: for any dataset that will be published, redistributed or used in training, knowing the origin page is not optional. Second, image — the full-size URL — is nullable on purpose. Google does not always expose it, and a guessed URL that 404s six months later is worse than an honest null.
query is required; country and lang shape which results Google serves, which matters more for image search than people expect — regional visual results differ substantially for the same term. max_results caps delivery and cost.
Each image is one row. thumbnail is always present and is what you fetch for a preview grid. image is the full-size URL when Google publishes it and null when it does not. source is the host and source_url the exact page — keep both: the host is for filtering and aggregation, the page URL is for attribution and for going back to the original context when you need licence information.
$0.0003 per delivered image ($0.30 per 1,000). Nothing delivered means nothing charged, and the $2 monthly free credit covers roughly 6,666 images here. Volume tiers take up to 30% off.
$0.0003 per delivered image, 30 cents per 1,000 — the cheapest per-row collector after keyword ideas. A 50,000-image training or reference set is $15 at list price. Building the same set with a headless browser scrolling Google Images would cost more in compute and bandwidth than that before anyone writes a parser.
Blocked pages and retries are not billed, and a query that genuinely returns nothing costs nothing.
Google has no public API that returns Google Images results. The Custom Search JSON API can be pointed at image search, but it queries a Programmable Search Engine rather than the public vertical, and its free daily quota is small enough that it is a prototyping tool rather than a dataset tool.
Bing's Image Search API was the standard answer for years and was retired, which pushed a lot of teams onto scraping whether they wanted to be there or not. Several general-purpose search APIs offer an image endpoint, and they are reasonable — they are also, mostly, doing the same thing this does, with a different bill attached.
The one genuine alternative worth naming is a licensed stock library API. If what you need is images you are cleared to use commercially, buy them. This collector answers a different question: what images exist on the public web for this term, and where did they come from.
Google Images is a lazy-loading, infinite-scroll surface, which pushes almost every home-grown scraper straight to Selenium or Playwright. That decision is where the cost comes from:
Query per class, keep source_url for every row, and you have a dataset with provenance instead of an unattributable folder of files.
Watch which images rank for your brand terms and which hosts they sit on — unauthorised use usually shows up here first.
Find candidate imagery for an article and go straight to the source page to check licensing before using anything.
See which product photos rank for a product term, and which retailers are using whose imagery.
Up to 100 images per run, pagination handled internally. Wider sets are more queries, parallel up to your plan's rate limit — 60 requests/minute on pay-as-you-go, up to 1,200 on the top tier.
Legally, image search deserves more care than text: search results are public, but the images themselves are almost always copyrighted by someone, and a thumbnail URL is not a licence. Collecting result metadata is one thing; redistributing, republishing or training on the underlying images is a separate question governed by the rights in each image and by the law where you operate. Keep source_url precisely so those questions stay answerable. This is not legal advice.
There is a free allowance rather than a free product: $2 of usage each month with no card, which at $0.0003 per image is about 6,666 delivered images. Free scrapers exist on GitHub and most of them drive a headless browser through infinite scroll, which is the expensive way to get the same rows.
No — it returns URLs and provenance, not files. Every row carries a thumbnail URL, the full-size URL when Google exposes it, and the page the image came from. Fetching the bytes is a separate step you control, which is deliberate: it keeps you in charge of storage and of the licensing decision.
$0.0003 per delivered image — 30 cents per 1,000 — with no subscription. Volume tiers take up to 30% off, and blocked pages and empty results are never billed.
Because Google does not always expose it. The field is nullable on purpose rather than filled with a guessed URL that would break later. The thumbnail and the source page are always there, and the source page is where the original lives.
Not automatically. The search result is public data; the image itself is usually someone's copyrighted work, and a URL is not a licence. That is exactly why every row keeps the source page — so you can check the terms at the origin before using anything. For guaranteed commercial rights, licensed stock is the right route.
No. It runs on the HTTP tier, which is why it is one of the cheapest collectors in the catalogue. Home-grown Google Images scrapers usually need a browser for the infinite scroll; this does not.
Yes. country and lang shape which results Google serves, and regional differences in image results for the same term are larger than most people expect.
Skip the marketing. QuantumProxies.io publishes a machine-readable site map, Markdown for every page, a free MCP server, and APIs billed from the same balance as your proxies.
Connect in one command: npx -y quantumproxies-mcp