Google Image Search Results & Metadata Scraper

Search Google Images for one or many queries and get one row per image, with an honest label on what each URL actually is. Google is tried first. If Google blocks or returns nothing, the default path returns explicitly labelled Bing/DuckDuckGo rows; set googleOnly=true for a strict Google-only contract that fails instead.

What this actor does

• Runs one or more Google Images queries in a single run.
• Exposes Google's own size, colour, type, file-type, time and usage-rights filters, plus country, language and SafeSearch.
• Captures the original image URL when Google exposes it, and tells you when it only has a thumbnail (imageUrlType, isCrawlerOnly, hasOriginal).
• Records the source page, source domain, site name, result position and page/scroll batch.
• Gives every row a stable imageId derived from the image URL or Google's result key.
• Writes a SUMMARY key-value record with per-query resultsRequested, resultsReturned, pagesFetched, stopReason and fallbackAttempts.

Google-first with labelled fallback

The default contract is labelled fallback: googleOnly is false and enableFallback is true. Google is always tried first. If Google is blocked or has no results, the actor tries Bing Images (plain HTTP) and DuckDuckGo Images, in that order. Every fallback row carries:

• engine - bing, duckduckgo, or google
• isFallback - true for every fallback row
• fallbackReason - why Google was not used, e.g. google_captcha

Fallback is never silent and never changes the engine value. A Bing row is always labelled Bing.

For a strict Google-only run, set googleOnly=true and enableFallback=false. If Google is blocked or has no results, the run fails with an explicit status instead of returning another engine.

Output fields

Every row keeps the legacy field names and adds provenance and completeness fields. Empty optional values are omitted; null and sentinel values are never written.

imageUrl is the best URL available, not a promise that it is the original. Check imageUrlType or hasOriginal before treating it as full resolution.

Input

At least one queries entry (or the legacy query string) is required. Apify's validator cannot express "one of", so the actor fails clearly when neither is present.

The legacy query and maxResultsPerQuery keys are read in code for saved tasks. There is no proxy field in the schema; the actor never requires a user-supplied proxy or API key.

Example input

Strict Google-only query:

{
  "queries": ["Alfons Mucha"],
  "maxResults": 20,
  "country": "US",
  "language": "en",
  "googleOnly": true
}

Labelled fallback allowed, with filters:

{
  "queries": ["sunset landscape", "modern architecture"],
  "maxResults": 100,
  "country": "DE",
  "language": "de",
  "imageSize": "large",
  "imageColor": "blue",
  "imageType": "photo",
  "safeSearch": "active",
  "googleOnly": false,
  "enableFallback": true
}

How to read a run

• A row with engine: "google" and imageUrlType: "original" is a source-site image URL.
• A row with engine: "google" and imageUrlType: "thumbnail" is Google's cached thumbnail; the original was not exposed for that card.
• A row with engine: "bing" is a labelled fallback row. Bing exposes original image URLs and source pages directly.
• The SUMMARY key-value record tells you how many rows were requested and returned per query, how many pages were read, why reading stopped, and which fallback engines were attempted.

Limits and honest notes

• Google Images typically runs out at roughly 300-400 results per query. Asking for 500 returns what Google has.
• Google sometimes exposes only thumbnails in the results DOM. The actor labels those rows thumbnail rather than pretending they are originals. It attempts the side panel for up to 20 cards per query to recover originals.
• A Google CAPTCHA or "unusual traffic" page is reported as google_captcha, not as zero results.
• fileType is applied by Google server-side. On Bing fallback it is applied to the image URL extension locally, so URLs without a recognisable extension are excluded when this filter is set.
• Bing's time filter has no "past hour" bucket; the hour option uses Bing's 24-hour window on fallback.
• The DuckDuckGo fallback reads its first image JSON batch only; it is a last resort after Bing. It has no size, colour, type, time or usage-rights facet, so it is skipped when one of those filters is set instead of returning rows that violate the filter.
• minWidth and minHeight are applied to every engine. Rows with no published dimensions are dropped when either bound is set.
• dominantColor and relatedSearches come from Bing's response. Google does not expose a reliable dominant colour in its search metadata; file size is not shipped because it would require downloading or probing every image.
• Usage-rights labels come from the search engine. They are not a legal guarantee; check the licence on contentUrl before reuse. Bing's licence buckets do not match Google's four labels one-to-one, so fallback rows use the nearest Bing bucket.
• New fields are additive. Existing field names (query, imageUrl, imageWidth, imageHeight, thumbnailUrl, thumbnailWidth, thumbnailHeight, title, origin, contentUrl) are preserved.

FAQ

Why did a Google-only run fail? Google returned a CAPTCHA, an error page, or no results. In Google-only mode that is a real failure and is reported as one. Set googleOnly to false and enableFallback to true if you want labelled Bing/DuckDuckGo rows instead.

Why is imageUrlType sometimes thumbnail? The results page did not expose the source image URL for that card. The actor will not label a Google thumbnail as an original.

Why are dimensions missing on some rows? The engine did not state them. The actor omits the field rather than writing 0 or guessing.

Does this download images? No. It returns URLs and metadata. Fetch the image files yourself.

Is this affiliated with Google? No. It reads public image search results as an independent third-party tool.

Can I use the same input as before? Yes for the common case: a saved task with query and maxResults still runs because the legacy keys are mapped in code. Batch users should move to queries and maxResults (per query).