How do I configure a media analysis widget?

Overview

A media analysis widget shows AI-read results from your photos or text submissions, checked against a Check you configure — for example, counting items in a shelf photo or confirming a display is present.

When to Use

Use this feature when:

You want an automatic read on photos or text submitted through a form — counting items, extracting a value, or scoring against a Check

You don't need true computer-vision accuracy (facing recognition, object detection, SKU matching) — this feature reads images with a language model, not a vision model built for that

Concepts

Media Analysis Widget – Shows the results of an AI read of your media (photos or text submissions) against a Check you configure

Check – What the AI is asked to do on each piece of media; choose one of five types. A Check that's already set up shows as "(configured)" in the dropdown

Uniform check – A yes/no check against a single condition in the photo, such as whether something matches an expected standard

Display / POSM present – A yes/no check for whether a point-of-sale display or material (POSM) is visible in the photo

Count items – Counts how many of something appear in the photo — for example, how many units of a product are on a shelf; configured using one or more Values

Read a value – Extracts a specific value from the photo or text, such as a price or a label reading; configured using one or more Values

Quality score – Scores the photo or submission against quality criteria you describe, instead of a yes/no result or a count

Value – A specific thing the AI should extract or count under a Count items or Read a value Check, given a name, a type (such as Number), and a description telling the AI what to look for; Number-type Values become ordinary columns you can sum or average elsewhere in Vantage (a KPI card, chart, or summary/pivot table)

Item Reference Images – Optional reference photos you can upload so the AI knows what a correct item looks like

Credits – Each media analysis job that completes successfully uses a credit; credit use is capped and tracked by your admin

Media Analysis Job – The background task that reads one piece of media against a Check; jobs run asynchronously, and the dashboard refreshes automatically every 15 seconds while jobs are in progress

Set Up Guide

Go to a dashboard and click Add Widget, then set the Title, Apply Dashboard Filters, Widget Type (Media Analysis), and Data Source as usual

Choose the media field to analyze — a photo or text column from your Data Source, such as Shelf Photo Url

Under "What should the AI check?", choose a Check — Uniform check, Display / POSM present, Count items, Read a value, or Quality score

For Count items or Read a value, click + Value to add one or more values — give each a name, a type (such as Number), and a description telling the AI what to look for

Optionally, upload Item reference images so the AI knows what a correct item looks like

Click Save & Apply

Best Practices

Stick to photo or text submissions — video and audio analysis exist as legacy options, but they don't produce reliable results today: video analysis only sends the model the file's URL (which it can't actually watch), and audio analysis returns a placeholder instead of a real transcript

Treat results as LLM-grade, not CV-grade — there's no true object detection or facing recognition behind this; the model is reading the photo and estimating, so spot-check anything used for a compliance or scoring decision

Give each Value a clear, specific description — it's literally the instruction the AI reads to decide what to extract or count, so a vague description gives vague results

Avoid saving the same Check multiple times in quick succession — each save can queue new analysis jobs, and resaving repeatedly can create duplicate jobs that use up extra credits

FAQs

Is media analysis real computer vision, like object detection?
No — it's a language model reading the photo (or text) against the rubric you set up. There's no true object detection, facing recognition, or SKU matching behind it, so treat results as LLM-grade, not CV-grade.

Can I use it to analyze videos or audio recordings?
Video and audio options exist but aren't reliable today — video analysis only sends the model the file's URL, which it can't open, and audio analysis returns a placeholder instead of a real transcript. Stick to photos and text submissions.

Does every analysis use up credits?
Yes — each job that completes successfully uses a credit, and credit use is capped and tracked by your admin.

Version Notes

Last Updated: Oct 2026