DEV Community

dodou
dodou

Posted on

Scrape Google Images Search Results with a SERP API (Python)

#python #api #images

Back when I needed product images for a comparison site, my first instinct was to parse Google Images' DOM. Two days later I had a pile of brittle selectors and a CAPTCHA habit. The fix was boring: an API that returns the image grid as JSON. Here's the shortest Python version I use.

The request

SerpBase's images endpoint is a POST with a JSON body — same shape as their web search, but it costs 2 credits instead of 1 because image parsing is heavier on their side.

import os
import requests

API_KEY = os.environ["SERPBASE_API_KEY"]

resp = requests.post(
    "https://api.serpbase.dev/google/images",
    headers={"X-API-Key": API_KEY, "Content-Type": "application/json"},
    json={"q": "mechanical keyboard", "hl": "en", "gl": "us", "page": 1},
    timeout=30,
)
resp.raise_for_status()

for img in resp.json()["images"][:10]:
    print(img["rank"], img["title"], img["image_url"])
    print("   source page:", img["link"])
Enter fullscreen mode Exit fullscreen mode

What comes back

The images array is the grid you see on Google Images, as rows. Per the docs (I re-checked today), three fields are always there and a few more show up when available:

Field Always? Use
rank ✅ Grid position — correlates with relevance
image_url ✅ The actual image file URL
link ✅ The page the image lives on
title optional Caption text
source / domain optional Publisher, for filtering by site quality

One honest caveat from the docs: some payloads only carry rank/link and skip normalized aliases like thumbnail — so code defensively with .get().

The workflow I settled on

Metadata first, download second. The script above pulls image URLs and source pages into a CSV; a separate step downloads only the ones that pass filters (min resolution via a HEAD check, domain allowlist, dedupe by URL hash). This keeps the credit bill low — you only pay for the search, not for hosting junk you'll throw away.

Pagination matters here: there's no num parameter, just page (1-based). Twenty pages of one keyword beats one page of twenty keywords when you're hunting a specific product shot.

Next step

Set your key (new accounts get 100 free searches, no card) and run it on a niche term. The full parameter list lives in the SerpBase /google/images endpoint docs. Your first useful filter will probably be domain — half the junk disappears once you blacklist aggregator sites.

Top comments (0)