# Vision and visual search

> Let shoppers upload a photo to find similar products; which models support it, how photos are handled and when they are deleted.

Source: https://shopassist.coec.dk/docs/ai/vision-and-visual-search · Updated 2026-10-01

With visual search, shoppers can upload a photo ("do you have something like this?") and get a carousel of similar products.

## Requirements [#requirements]

* **Settings → Commerce → Visual search** is on.
* The store's AI model understands images. Managed AI does; with your own key, the connection test's `vision` check must pass (`vision_unsupported` switches visual search off).

## How it works [#how]

1. The shopper attaches up to 3 photos per conversation. Photos are resized and their metadata (EXIF, including location) is removed before storage.
2. A separate vision call describes the product in the photo (type, colours, material, style, visible text). The model that answers the shopper never sees the image itself, only this sanitized description.
3. ShopAssist searches your catalog with the description and shows the closest products. The assistant then refines with follow-up questions.

The photo is used only to find products. The vision step is instructed not to identify people.

## Retention [#retention]

Uploaded photos are deleted after 24 hours. Later in the conversation the photo appears only as its text description.

## Limits [#limits]

Shoppers can upload 10 photos per hour per chat session and 3 per conversation. Photo analysis is part of the reply; there is no extra charge per image.