Street View Semantic Search — find anything in Google Street View with AI

Search real-world streets with natural language. Give this actor a location, a radius, and a plain-English prompt — it scans every Google Street View panorama in the area and scores each image on how well it matches your prompt, using CLIP AI image understanding. Results stream into your dataset live as the scan progresses.

No API keys, no setup, no scraping infrastructure. Type what you're looking for and get scored, geotagged street imagery back in minutes.

"red storefront"        → every shopfront in the area, reddest first
"outdoor cafe seating"  → pavement terraces, ranked
"construction site"     → active building works with coordinates
"street art mural"      → murals and graffiti walls, scored

What can you find?

• 🏪 Retail & real estate — storefronts, vacant units, for-sale signs, building conditions
• 🏗️ Construction & infrastructure — scaffolding, roadworks, cranes, utility works
• 🌳 Urban planning & environment — green spaces, bike lanes, street furniture, accessibility ramps
• 📸 Location scouting — film/photo backdrops: "cobblestone alley", "neon signs at dusk"
• 🚗 Mobility & parking — EV chargers, parking lots, bus stops, loading bays
• 🎯 Lead generation — restaurants with terraces, shops with awnings, businesses with billboards

Any question you could answer by walking the streets and looking — this actor answers by looking for you.

How it works

1. You provide a location (address or coordinates), search radius, and a text prompt
2. Panorama discovery — the actor sweeps the area and finds every available Street View panorama
3. Image capture — it photographs each panorama from multiple compass headings
4. AI ranking — a CLIP vision model scores every image against your prompt (semantic matching, not keywords: "place to sit outside" finds benches and terraces)
5. You receive every image as a dataset item — streamed live during the run — each with a match rating (match, ambiguous, or no_match), coordinates, capture date, and a download link

Input

Field	Default	Description
location	—	Address ("Camden High Street, London") or coordinates ("51.5033,-0.1196")
searchPrompt	—	What to look for, in plain English
radiusMeters	250	Search radius (10–5000 m)
maxImages	400	Maximum images to capture (= maximum results)
headings	[0, 90, 180, 270]	Camera directions per panorama
newerThan	—	Only imagery captured after this date, e.g. "2024-01" — older panoramas are skipped before download
gridSpacingMeters	15	Discovery density — smaller finds more panoramas

Output

One dataset item per image, streamed in as it's captured. Every image carries an AI match rating — match, ambiguous, or no_match — so you can filter straight down to the hits.

{
    "match": "match",
    "panoId": "CAoSF0NJSE0...",
    "lat": 51.50335,
    "lng": -0.11989,
    "heading": 0,
    "date": "2023-06",
    "imageUrl": "https://api.apify.com/v2/key-value-stores/.../result-001-....jpg"
}

Export to JSON, CSV, or Excel; images are stored with the run and downloadable via imageUrl.

Why this actor?

• ✅ Zero setup — no Google API key, no billing account, no quota management
• ✅ Semantic search, not keyword search — CLIP understands concepts, colors, objects, scenes
• ✅ Complete results — every scanned image returned with its match rating and location, nothing hidden
• ✅ Geotagged — every result carries exact coordinates and Street View capture date
• ✅ Official imagery — sourced from the Google Street View API, no fragile scraping

FAQ

How many images will I get? Up to maxImages. A 250 m urban radius typically holds 50–150 panoramas; each is captured from the headings you choose (4 by default).

Does it work everywhere? Anywhere Google Street View has coverage — most cities and roads worldwide. No coverage = zero results, and you're only charged for images produced.

What do the match ratings mean? match means the image is a confident visual match for your prompt; ambiguous is borderline and worth a look; no_match means the image probably doesn't contain what you asked for. Filter the dataset on the match field to keep only the levels you want.

Can I search for multiple things? One prompt per run. Kick off parallel runs for multiple prompts over the same area.

How fresh are the images? Each result includes its Street View capture date (e.g. "2023-06"). Set newerThan (e.g. "2024-01") to skip older panoramas entirely — you're only charged for images that pass the filter. Note: areas Google last photographed before your cutoff will return no results.

---

Street View scraper · semantic image search · CLIP visual search · Google Street View images by location · AI street imagery analysis