Skip to content
FakeGuardAI
Verify a productCreate account

Counterfeit detection · gold jewellery

See what the AI saw.

FakeGuard AI estimates the counterfeit risk of a gold jewellery item from a photo, and shows you the areas that weighed on its verdict. A decision aid, not a certificate.

Verify a productAnalyse an itemRead the model's measurements
Test-set specimen f_img_559
1

Drag across the image to move the loupe Loupe: arrow keys to move it, Enter to show the whole attention map.

Test-set specimen
f_img_559
Source label
counterfeit
Model verdict
High risk
Estimated probability of counterfeit
> 99%

Photo: GoldNet, Z. Raisi — CC BY 4.0

Three verdicts. Never “certified authentic”.

A photo cannot prove that an object is authentic. When the model is not sure enough, it says so, and you can ask for a human look.

  1. Low risk · Low risk · Low risk · Low risk ·

    Low risk

    The model finds no sign of counterfeiting. This is not a guarantee of authenticity.

    r_img_732
    Source label : authentic
    Estimated probability of counterfeit : < 1%
  2. Inconclusive · Inconclusive · Inconclusive ·

    Inconclusive

    The model does not decide: the photo is outside its scope, or the score is too close to the limit.

    r_img_776
    Source label : authentic
    Estimated probability of counterfeit : 57%
  3. High risk · High risk · High risk · High risk ·

    High risk

    The model finds signs consistent with a counterfeit. Have the item checked before any decision.

    f_img_559
    Source label : counterfeit
    Estimated probability of counterfeit : > 99%

Measured, not promised.

These figures come from the test set: 321 items the model never saw during training. In brackets, the 95% confidence interval.

79%

of counterfeits flagged “high risk”

[73% – 85%]

96%[93% – 99%]

of “high risk” verdicts were correct

4%[1% – 8%]

of counterfeits wrongly rated “low risk”

17%[12% – 21%]

of analyses left inconclusive

The full model card

How an analysis unfolds

  1. 01

    Take the photo

    One to four sharp photos, item centred, no filter or retouching.

  2. 02

    Photo check

    A blurry, too dark or too small image is rejected before analysis, with the reason.

  3. 03

    Calibrated verdict

    The model estimates a probability of counterfeit, then decides — or abstains.

  4. 04

    Human look

    When in doubt, request a reviewer's opinion from the report.

What FakeGuard AI does not claim

We would rather write down the model's limits than let you discover them.

  1. 01The result is a statistical estimate from a photo. It replaces neither a metal assay nor a professional's opinion.
  2. 02The model learned from a single public dataset: GoldNet, 2,127 items.
  3. 03The source does not document how the authentic / counterfeit labels were established.
  4. 04The items come mostly from Iran and the Gulf market: reliability on other styles has not been measured.
  5. 05Each item has a single photo in the dataset: combining several photos has not been validated experimentally.
  6. 06The filter that rejects off-topic photos was only checked on a small informal probe.
  7. 07The setting of the photos is partly enough to separate the two classes: the model may have learned context cues rather than material ones.

Start with one item.

Create an account, drop a photo, read the report.

Analyse an item