Google Lens Pro

Google Lens Pro Agent Skill

This skill allows an AI agent to execute reverse image searches, extract optical character recognition (OCR) text, detect visual entities, and find shopping matches/prices via Google Lens.

Prerequisites & Installation

If lens is not yet installed in the environment, install the package and download the stealth Chromium browser binary:

pip install google-lens-pro
patchright install chromium

Note: Fast-path OCR mode (--ocr-only) only requires the Python package and works immediately without downloading Chromium binaries.

Quick Start Recipes

1. Reverse Image Search & Visual Intelligence (Default Mode)

To find matching products, web sources, normalized pricing analytics, and multimodal attribute analysis:

google-lens search "<image-url-or-path>" --json-output

Note: By default, google-lens search automatically executes e-commerce price normalization (--enrich) and deep multimodal attribute analysis (--analyze) via Gemini 3.8 Flash if GEMINI_API_KEY is present.

Agent In-Context Fallback Strategy: If results.analysis is null (because GEMINI_API_KEY is not configured in the host environment), you (the AI Agent) should inspect the query image and the returned visual_matches directly within your own conversation context to deduce product attributes (brand, model silhouette, materials, condition, estimated MSRP, authenticity cues). This guarantees zero friction for users without requiring additional API keys.

2. Nano Banana Pro 8K AI Studio Packshot Generation

To synthesize an ultra-crisp 8K commercial catalog packshot on a pristine white background from a reference image:

google-lens search "<image-url-or-path>" --studio --studio-output ./packshot.png

You can optionally provide a custom prompt:

google-lens search "<image-url-or-path>" --studio --studio-prompt "8K studio catalog shot on black brushed granite with soft dramatic rim lighting"

3. Fast-Path OCR & Object Detection (Sub-second, No CAPTCHA)

For extracting text or bounding boxes from an image (uses Chromium Protobuf endpoint, zero browser overhead):

google-lens search "<image-url-or-path>" --ocr-only --json-output

4. E-Commerce & Resale Arbitrage Intelligence (Pro)

To resolve clean canonical URLs, normalized price ranges (min/max/average), semantic match relevance, and identify the lowest-priced verified seller ("Best Deal"):

google-lens search "<image-url-or-path>" --enrich --json-output

Pro enrichment categorizes candidates against the identified target_product using Gemini 3.8 Flash into universal semantic classes:

• exact_match: Exact physical product match (eligible for price comparison and Best Deal).
• similar: Same silhouette, competing brand, or variant edition.
• reference: Blog reviews, worksheets, wallpapers, or editorial context.
• unrelated: Incompatible accessories, third-party parts, or SERP noise.

To export enriched commerce data directly to clean JSON:

google-lens search "<image-url-or-path>" --export-json deals.json

5. Real-Time Token & Cost Accounting

Every Gemini API invocation (Gemini 3.8 Flash and Nano Banana Pro) is automatically tracked using the built-in gemini-cost-calculator. When calling via CLI or SDK, results.cost returns exact token usage and financial cost in USD:

"cost": {
  "model": "gemini-3.8-flash",
  "calls_count": 1,
  "tokens": { "prompt": 450, "output": 180, "total": 630 },
  "cost_usd": { "total": 0.001013 }
}

6. License Activation & Setup (Pro)

Pro features (market pricing intelligence, clean URLs, JSON export, studio packs) require an active Polar.sh license.

Purchase a License (Browser):

google-lens buy pro
# Or specify a plan:
google-lens buy --plan annual

Opens the Polar.sh checkout page directly in your browser to choose a plan (Lifetime, Annual, or Monthly).

Activate in Terminal:

google-lens pro activate "<your-polar-license-key>"
# Or run interactively:
google-lens pro activate
# Or set environment variable:
export LENS_LICENSE_KEY="<your-polar-license-key>"

Check License Status:

google-lens pro status

Deactivate on this Machine:

google-lens pro deactivate

7. Google AI Studio Key & Billing Tier (Optional)

Google Lens Scraper does not require any Gemini API key for regular searches or product attributes—those are deduced natively from Lens metadata with zero friction.

To synthesize 8K AI studio packshots (--studio) or run deep multimodal attributes, configure your key:

google-lens setup-ai --key "<your-gemini-api-key>"
# Or set environment variable:
export GEMINI_API_KEY="<your-gemini-api-key>"

Configure your billing tier (unknown, free, or paid) for exact USD cost accounting:

google-lens setup-ai --tier paid
# Or set environment variable:
export GEMINI_BILLING_TIER="paid"

8. Session Status & Google Authentication

Google Lens requires an authenticated Google session for complete visual matches and to avoid /sorry/index rate limits:

google-lens status

If unauthenticated and running in an interactive desktop environment:

google-lens login

For headless/cloud/CI environments, pass session state via environment variable:

export LENS_STORAGE_STATE_JSON="<base64-or-raw-json-storage-state>"

Bundled Helper Scripts

For programmatic workflows and batch operations, use the bundled scripts located in scripts/. These scripts include PEP 723 inline dependency metadata.

Single Image Search & Filtering (scripts/search_image.py)

Search an image with built-in filtering by price, result count, or mode:

# Basic search with JSON output
python3 scripts/search_image.py "https://example.com/photo.jpg" --json

# Filter visual matches (e.g. max 5 results)
python3 scripts/search_image.py "path/to/image.png" --max-results 5 --json

# Fast OCR extraction
python3 scripts/search_image.py "path/to/receipt.jpg" --ocr-only --json

Batch Image Processing (scripts/batch_search.py)

Process a directory of images or a list of URLs asynchronously:

# Process all images in a directory
python3 scripts/batch_search.py --dir ./images/ --output ./batch_results.json

# Process list of URLs from file
python3 scripts/batch_search.py --urls-file ./image_urls.txt --output ./batch_results.json --concurrency 3

Interpreting Output

When --json-output is used, the JSON payload follows this structure:

{
  "query": "https://example.com/item.jpg",
  "ocr_text": "Detected text from the image...",
  "knowledge_graph": {
    "title": "Identified Object / Entity Name",
    "subtitle": "Classification (e.g. Footwear / Plant)",
    "description": "Brief encyclopedic description",
    "website": "https://..."
  },
  "commerce": {
    "summary": {
      "target_product": "Nike Free RN Flyknit 2017",
      "min_price": 120.0,
      "max_price": 120.0,
      "avg_price": 120.0,
      "currency": "USD",
      "total_matches": 69,
      "total_priced_matches": 1
    },
    "items": [
      {
        "title": "Nike Free RN Flyknit 2017",
        "direct_url": "https://www.goat.com/sneakers/free-rn-flyknit-2017-880843-006",
        "match_score": 97,
        "relevance": "exact_match",
        "relevance_reason": "Direct marketplace listing for identical silhouette.",
        "page_type": "product",
        "price": { "raw": "120.00 USD", "amount": 120.0, "currency": "USD" },
        "merchant_name": "GOAT"
      }
    ]
  },
  "visual_matches": [
    {
      "title": "Product Title",
      "link": "https://store.com/item",
      "source": "Store Name",
      "price": "$99.00",
      "thumbnail": "https://..."
    }
  ],
  "detected_objects": [
    {
      "id": "item_1",
      "bounding_box": {
        "center_x": 0.5,
        "center_y": 0.5,
        "width": 0.3,
        "height": 0.4
      }
    }
  ]
}

Troubleshooting & Best Practices

1. Bot Clearance (HTTP 429 / CAPTCHA): If visual match requests are rate-limited, switch to --ocr-only (guaranteed zero CAPTCHAs) or authenticate via google-lens login or LENS_STORAGE_STATE_JSON.
2. Local Images: Local file paths (.png, .jpg, .jpeg, .webp, .bmp, .gif) are automatically uploaded to Google's fast-upload endpoint.
3. Headless Execution: Searches run headless by default. In debugging environments, pass --no-headless to inspect the browser.

For complete CLI flags, configuration parameters, and detailed JSON schemas, see references/REFERENCE.md.