Skip to content
vviinn DocsDocs Beta

Visual Search

Visual Search lets a shopper search with a photo instead of words — a picture of a product, a screenshot, or a photo of a barcode or printed label. Three detection methods run at once and their results are merged into a single ranked response.

How it works

Section titled “How it works”
flowchart TD
    A[Shopper uploads or takes a photo] --> B{Detection}
    B --> C[Visual similarity match]
    B --> D[Barcode / QR lookup]
    B --> E[Text recognition\nOCR]
    C --> F[Merge results]
    D --> F
    E --> F
    F --> G[Filters applied]
    G --> H[Results shown to shopper]

What it does

Section titled “What it does”

  • 🖼️ Visual similarity

    The image is matched against your catalog by visual resemblance — the primary way most visual searches are answered.

  • 🏷️ Barcode reading

    If the photo contains a barcode or QR code, it’s decoded and looked up directly against your catalog’s GTIN/MPN identifiers. A single confident match can take the shopper straight to that product page.

  • 🔤 Text recognition (OCR)

    Text visible in the photo — a label, a printed name — is read and used to run a search alongside the visual match.

  • ✂️ Region selection

    Shoppers (or the widget) can crop to a specific area of the photo to search just that part of the image, and results can be filtered by which detection method produced them (visual, barcode, or text).

Configuring in Panel

Section titled “Configuring in Panel”

In Panel, open Site → Campaigns, and open or create a Visual Search campaign.

General settings — Name, Active, Default Campaign toggles, same as other campaign types.

Methods

  • Semantic Search — turn on and weight the AI models used for visual matching (e.g. Volcanic Water Net, Berry Punch Net); a Net must first be turned on for the site under AI methods.
  • Color — turn on color-aware matching if your site has a color-capable Net (e.g. Kiba Net) activated.

Influencers

  • OCR — toggle text recognition on or off for this campaign.
  • Barcode reader — toggle barcode/QR decoding on or off.
  • Only Buyable Products — restrict results to in-stock products.

Interface controls

  • Preferred Image — which product image to use when comparing against catalog products.
  • maximum results — cap on returned results.
  • Object Detection (available on advanced-configuration plans) — turn object detection on, cap Maximum Objects detected per photo, choose which object classes are relevant (default set, or manual if you want to specify your own), and set a confidence threshold below which a detected object is ignored.
  • Filters — same filter configuration panel as Text Search: choose which catalog attributes are filterable, label them, mark some as Primary, and set filter order.

Products — restrict the campaign to specific categories via Assigned Products, or leave unrestricted.

With item grouping on, matching variants of the same product are grouped into a single result.

Boosting is not currently available for Visual Search.

MCP Tool

Section titled “MCP Tool”

Agents can call this service directly as the visual_search MCP tool, which wraps the Visual-Search Products REST operation — its request and response shapes match that operation’s schema.