> ## Documentation Index
> Fetch the complete documentation index at: https://docs.highailabs.com/llms.txt
> Use this file to discover all available pages before exploring further.
# AI Label Scanner
> Scan cannabis product labels with AI-powered vision to extract terpene profiles, cannabinoid data, and get instant High Family classification.
## Overview
The Label Scanner uses **Google Gemini 3 Flash** vision AI to read cannabis product labels -- dispensary stickers, COA (Certificate of Analysis) labels, and retail packaging -- and extract detailed chemical data. In seconds, you get a complete terpene profile extraction, cannabinoid breakdown, and classification into one of the 6 High Families.
This bridges the gap between what is printed on a label and what it actually means for your experience.
The in-app Label Scanner is a **Pro feature**. Because every scan runs live AI vision, it requires a signed-in account with an active Pro subscription (\$12.99/month). See [What's in Pro](/help/plans/whats-in-pro).
## Key Features
Gemini 3 Flash reads strain name, brand, THC/CBD percentages, and individual terpene amounts directly from label photos.
Extracted terpene profiles are matched to one of 6 High Families with a confidence score, telling you what kind of experience to expect.
Scan one label or up to 4 images at once. Batch mode is ideal for curved containers or multi-sided packaging.
If the scanned strain exists in the 16,000+ strain database, you get an automatic link to its full profile page.
If a scanned strain is not in the database, you can opt in to submit it for automatic verification and addition.
Images are processed and immediately discarded. IP addresses are hashed. No personal data is stored from anonymous scans.
## How to Scan a Label
Navigate to the Label Scanner from the Quick Actions on the Home dashboard, or from the **You** tab.
Take a photo using your camera or select an existing image from your photo library. For batch mode, add up to 4 images.
In batch mode, optionally label each image (e.g., "front," "back," "COA") to help the AI prioritize data sources.
Tap "Scan" to send the image(s) to the AI. An animated processing state shows while the analysis runs.
Results display the extracted data, High Family classification with confidence score, and a link to the strain profile if found in the database.
## What Gets Extracted
The AI attempts to extract the following from each label image:
| Data Point | Description |
| ---------------------- | --------------------------------------------- |
| **Strain Name** | The product or cultivar name |
| **Brand Name** | The producer or brand (e.g., Cresco, Cookies) |
| **Product Type** | Flower, concentrate, edible, vape, etc. |
| **THC %** | Total THC content |
| **CBD %** | Total CBD content |
| **Other Cannabinoids** | CBG, CBN, THCA, CBDA if listed |
| **Terpene Profile** | Individual terpene names and percentages |
| **Label Type** | Identified as COA, retail, or dispensary tag |
## Classification Tiers
The accuracy of the High Family classification depends on how much data the AI can extract:
| Tier | Data Available | Accuracy | Description |
| ---------- | ------------------------------ | -------- | --------------------------------------------------------------- |
| **Tier 1** | Complete terpene percentages | 80-95% | Direct cluster matching via terpene ratios. Best results. |
| **Tier 2** | Dominant terpene + strain type | 60-80% | Maps to the most likely family based on the primary terpene. |
| **Tier 3** | Strain type + effects | 40-60% | Uses type-based heuristics when terpene data is limited. |
| **Tier 4** | Strain type only | 20-40% | Fallback classification. Low confidence indicator is displayed. |
**COA labels produce the best results** because they contain exact terpene percentages. Retail packaging with only THC/CBD listed will yield lower-confidence classifications.
## Batch Scanning
Batch mode processes 1-4 images and intelligently merges the results:
* **Terpenes** -- Takes the maximum value for each terpene across all images
* **Cannabinoids** -- Prefers COA data over retail label data
* **Strain Name** -- Uses the extraction with the highest confidence score
This is especially useful for:
* **Curved glass containers** where no single photo captures the entire label
* **Products with separate COA stickers** on different sides
* **Pre-roll tins** with strain info on the lid and terpene data on the bottom
## Photography Tips
**Good lighting** is critical. Natural light works best -- avoid direct glare on glossy labels.
**Fill the frame** with the label text. The closer and sharper the text, the better the extraction accuracy.
**COA stickers are gold.** If your product has a Certificate of Analysis sticker, scan that first. It contains the most precise terpene and cannabinoid data.
For **curved containers**, use batch mode with 2-4 close-up sections rather than one wide-angle shot.
## High Families
The scanner classifies products into one of 6 families based on terpene data. These are rooted in a [2025 peer-reviewed German chemovar study](https://thisiswhyimhigh.com/learn/german-chemovar-study-high-families-science) that found 9 terpenes explain 86% of chemical variation in cannabis:
| Family | Experience | Key Terpenes |
| ------------------ | ------------------------------ | ----------------------- |
| **Relaxing High** | Calm, sedating, body-focused | Myrcene, Linalool |
| **Uplifting High** | Energizing, mood-boosting | Limonene, Pinene |
| **Energetic High** | Focused, active, cerebral | Terpinolene, Pinene |
| **Relieving High** | Anti-inflammatory, therapeutic | Caryophyllene, Humulene |
| **Balancing High** | Versatile, mild, even-keeled | Mixed terpene profile |
| **Entourage High** | Complex, full-bodied, layered | Diverse terpene spread |
## Edge Cases
The scanner handles imperfect conditions gracefully:
| Situation | What Happens |
| -------------------------- | ------------------------------------------------------------------------ |
| **Not a cannabis label** | Friendly message explaining what types of images work |
| **Partial data extracted** | Warning banner shown; classification still attempted with available data |
| **Low confidence result** | Classification shown with a disclaimer about accuracy |
| **Strain not in database** | Option to submit it for automated verification and addition |
| **Image too blurry** | Prompt to retake the photo with tips |
## Privacy & Data
* **Images are never stored.** They are processed in memory and immediately discarded after extraction.
* **IP addresses are hashed** using SHA-256 before any analytics storage. The original IP cannot be recovered.
* **Website scans are anonymous.** No user account is required and no personal data is collected.
* **Mobile scans** include your user ID for scan history features.
* **Device data** (device type, browser, general location) is collected for analytics only.
## Performance
| Specification | Value |
| --------------------- | ------------------------------------------------------------- |
| **Processing Time** | 2-4 seconds typical |
| **Max Image Size** | 10 MB (\~7.5 MB after base64 encoding) |
| **Supported Formats** | JPEG, PNG, WebP, HEIC |
| **Auto-Resize** | Images larger than 1920px are resized on device before upload |
| **Rate Limit** | Rate limited per user (configurable) |