fastai search_images_bing without a Bing key

The function, before and after

This is search_images_bing in fastbook's utils.py, lines 54 to 61 at commit 971f0289c, as it is in the repo:

def search_images_bing(key, term, min_sz=128, max_images=150):    
     params = {'q':term, 'count':max_images, 'minHeight':min_sz, 'minWidth':min_sz, 'imageType':'photo'}
     headers = {"Ocp-Apim-Subscription-Key":key}
     search_url = "https://api.bing.microsoft.com/v7.0/images/search"
     response = requests.get(search_url, headers=headers, params=params)
     response.raise_for_status()
     search_results = response.json()
     return L(search_results['value'])

And with the one line changed:

def search_images_bing(key, term, min_sz=128, max_images=150):    
     params = {'q':term, 'count':max_images, 'minHeight':min_sz, 'minWidth':min_sz, 'imageType':'photo'}
     headers = {"Ocp-Apim-Subscription-Key":key}
     search_url = "https://api.specstream.ai/v7.0/images/search"
     response = requests.get(search_url, headers=headers, params=params)
     response.raise_for_status()
     search_results = response.json()
     return L(search_results['value'])

The params, the header, raise_for_status() and the return stay. The same function, with the same search_url line, is in fastai/course20's fastbook/__init__.py, and changes the same way.

Call it with max_images=100

The function sends max_images as count, and its default is 150, the most Bing took. SpecStream returns up to 100 images a call, so a call with 150 gets 400. Pass max_images=100, or set the default to 100 in the def line.

results = search_images_bing(key, 'grizzly bear', max_images=100)
ims = results.attrgot('contentUrl')
len(ims)  # 100

The book's own function returned 100 images

The book's own function, with that one line changed, returned 100 images on November 10, 2026, each with contentUrl, called as search_images_bing(key, "grizzly bear", max_images=100) with fastcore 2.2.41 and requests 2.34.2. download_images(dest, urls=results.attrgot('contentUrl')) then works as the book wrote it.

If your notebook calls search_images_serpapi

Today's fastbook calls search_images_serpapi in chapter 2. The function is in utils.py, lines 27 to 45 at commit 971f0289c, and one value in it changes, the url:

# before
url,params = 'https://serpapi.com/search', dict(engine=engine, q=term, api_key=key)

# after
url,params = 'https://api.specstream.ai/serpapi/search', dict(engine=engine, q=term, api_key=key)

Call it with engine='google_images' and put a SpecStream key in key or in the SERPAPI_KEY environment variable. The function's default engine, bing_images, gets 400 today.

urls = search_images_serpapi('grizzly bear', engine='google_images', key=key, max_images=30)
len(urls)  # 30

The book's own function, with that one value changed, followed serpapi_pagination.next twice and returned 30 image addresses on November 10, 2026. Each page it asks for is one search of 10 images at $0.005, so 150 images are 15 searches, $0.075.

$0.059 for 100 images, with no plan

An image search of up to 10 results costs $0.005, the price of a web search, and each started block of 10 more adds $0.006. The book's call for 100 images is $0.059, so the chapter's three bear types cost $0.177. The $10 of credit a new account gets covers 169 calls of 100 images. There is no plan and no daily limit, credit you buy never expires, and a call stopped by a 429 is not charged.

You decide what is kept, on each key

Create the key with zero_retention and a search made with it keeps no request text and no answer after the reply. spend_cap_cents is the most the key can spend, which suits a key pasted into a notebook.

curl -X POST https://api.specstream.ai/v1/account/keys \
  -H "Authorization: Bearer $SPECSTREAM_API_KEY" -H "Content-Type: application/json" \
  -d '{"name": "notebook", "scopes": ["account:read", "finds:spend"], "spend_cap_cents": 1000, "zero_retention": true}'

Images, web, news and pages on one key

SpecStream is the live web data engine. The key in your notebook also runs web search and news through GET /v1/web-search, scoring 97.9 F-score on all 1,000 SimpleQA Verified questions and 64.0 item F1 on all 200 WideSearch tasks; the method is on the benchmarks page. It reads pages with the Extract API at $0.80 per 1,000 pages, 0.908 F1 on WCXB, and Deep Search returns every match for a request in plain words as structured rows, at 5¢, 15¢ or $1 a row.

Every parameter and field of Bing's image call is in the Bing Image Search API migration guide, and every image API SpecStream answers is on the image search API page.

SerpApi, in today's fastbook

fastbook's chapter 2 now says Microsoft has retired the Bing Image Search API the book relied on, and uses a new function, search_images_serpapi. SerpApi's free plan gives 250 searches a month, and each page of 35 Bing results is one search, so a 150 image query costs 5. Starter is $25 a month for 1,000 searches. SpecStream answers both functions: search_images_bing at $0.059 for 100 images, and search_images_serpapi, with its url changed and engine='google_images', at $0.005 for each page of 10.

Grounding with Bing Search

Microsoft's named replacement, a tool for Azure AI Agents at $14 per 1,000 transactions. Microsoft says developers and end users don't have access to the raw content it returns, so there is no value[] and no contentUrl for download_images. SpecStream returns both.

Google Custom Search JSON API

Its image search costs $5 per 1,000 queries, but the API is closed to new customers and Google discontinues it on January 1, 2027. SpecStream answers Google's image call too, on the same key.