Google Images

Google Images through HasData

One tool, hasdata_google_images_images_getImageSearchResults. Only q is required, and one call returns a hundred results rather than ten.

Cost is per call rather than per result, so the cheapest shape here is one wide call. ijn is the page number and it starts at zero, which makes 1 the second hundred, not the first.

Two different URLs per result

original points at the file on whoever hosts it, so it carries none of Google's availability and can be dead, hotlink-blocked or slow. Fetch it defensively and fall back to thumbnail, which is served from Google's CDN as a URL rather than embedded data.

originalWidth and originalHeight describe the file behind original. Filtering by size locally means reading those rather than measuring whatever you displayed.

Filters fail quietly

tbs is Google's own filter string and several filters combine with commas, such as isz:l,ic:gray,itp:photo. An unrecognised value is ignored rather than rejected, and a filter that did nothing looks identical to a filter that matched everything. Check that the result set actually narrowed before trusting a tbs string assembled by hand.

The useful ones are isz:l, isz:m and isz:i for size, isz:lt,islt:qsvga upward for a floor, ic:color and ic:gray, ic:specific,isc:red for a named colour, and itp:photo, itp:face, itp:clipart, itp:lineart and itp:animated for type.

Locale is three separate axes

domain is which Google answers, gl is where the search comes from, and hl is the interface language. They move the results less than people expect, so change one at a time and compare.

Nothing here licenses an image. The results say where a file sits, not whether it can be used.