In short: AI Image Feature Detection spots the small visual details that separate your genuine products from lookalikes and counterfeits, such as packaging, logos and stitching, and flags them on each post so they can be reviewed and used in Auto-Moderation rules.
Understanding AI Image Feature Detection
Counterfeits are getting harder to spot by eye. The give-aways are often small: a misspelled logo, cheap packaging, stitching that doesn't match the original. AI Image Feature Detection finds these details in post images automatically, so the most convincing fakes don't slip through.
Unlike a generic image model, it's trained on your brand. It learns the fine visual traits of your genuine products and of known infringements, which makes it especially effective in industries like fashion, beauty and consumer goods, where products can look deceptively similar.
Note: AI Image Feature Detection needs to be trained on your products before it can detect anything. To set it up for your Program, speak to your Account Manager.
1. What it detects
It looks past surface-level appearance to the details that give an infringement away.
Product lookalikes: packaging nuances, construction methods and finishing details, such as stitching patterns or cap designs.
Authentic vs. infringing products: subtle design features, rather than overall appearance alone.
Borderline posts: extra visual evidence that raises confidence where a post is otherwise hard to classify.
Near-identical products: items that share the same form but differ in specific design execution.
Learn more: Logo Detection · AI Product Detection
2. How it works
Because the model is custom-trained on your brand, it recognises exactly the traits you care about.
Brand Guide: you provide a guide with examples of your brand's genuine traits and of known infringing features.
Data sourcing: a dataset of relevant images is sourced from the platform.
Manual labelling and training: experts label those images to train the model, so it learns to recognise exactly the examples you provided.
How detections are graded
Each feature is given a type, which tells Zeal how much weight to give it.
Type | What it means | Example (Solène) |
Info | A neutral or reference visual | The official Solène gift box |
Logo | A recognisable brand or trademark image | The Solène wordmark |
Suspicious Feature | Might indicate an issue | Uneven stitching |
Highly Suspicious | A strong indicator of infringement, but not definitive | A cheap dust bag, a misspelled logo |
Obvious Counterfeit | Clearly identifies a fake product | A colourway Solène has never produced |
Learn more: Image Feature Configuration
3. Reading detections on a post
Open a post to see exactly which features were found, and where.
On the image: each detected feature is marked with a green label, pointing to where it appears.
This Image / All Images: switch between the detections on the image you're viewing and those across all of the post's images.
Originally identified from another post: the feature was first found on a different post. Click Go To Post to open it.
4. Finding posts by image feature
Once your model is trained and live, you can pull up every post where a feature was detected. Use the Reasons filter in any Explore view and select the image feature you want to see.
Learn more: Reasons Filter · Explore View
5. Using detections in Auto-Moderation
Detected features can drive your Auto-Moderation rules, so posts are labelled automatically as soon as the details appear.
For example: if an image shows Dust Bag - Cheap and not Box - Solène, label the post as a potential counterfeit.
Learn more: Building Auto-Moderation Rules · Image Feature Configuration
Quick reference
If you want to know… | Look at… |
How to get AI Image Feature Detection set up for your brand | Your Account Manager |
How a detection is weighted | How detections are graded |
What was detected on a post | 3. Reading detections on a post |
Which posts have a specific feature | 4. Finding posts by image feature |
How to label posts automatically from detections | |
How to add or edit a feature |