Method
A resemblance reading, with its reasoning shown.
01
We read the packaging
Your photos go through a vision model that transcribes printed text — brand, product name, claims, size — and describes the packaging: container shape, cap, colour palette, typography, layout.
02
We turn that into two fingerprints
One fingerprint captures how the pack looks. A second captures the wording on it. Both are stored as vectors so any two products can be compared numerically.
03
We compare within the product type
A serum is compared with serums first. Products from other categories still appear, but under 'Related', so a cleanser that happens to share a bottle shape doesn't crowd out the real matches.
04
We score and explain
Sixty percent of the score comes from visual resemblance, forty from the label text, with a small boost when brand or product names overlap. Every match lists the specific reasons behind its number.
What this is not
- Not an authenticity check. A high score means two products look alike. Look-alike packaging is often perfectly legal.
- Not a formulation comparison. We read what is printed on the pack, not what is inside it.
- Not exhaustive. We only compare against products people have submitted. A low score can simply mean the look-alikes are not in the index yet.
How we make money — and why it can’t move a score
Some result pages carry a “Buy the original” link to a retailer, and we may earn a commission if you buy through it. Brands can also pay us to be alerted when look-alikes of their products are submitted.
Neither payment can change a similarity score, suppress a match, or decide what gets flagged. Scores are produced by the model and the formula described above, before any commercial link is attached. If a paying brand’s own product resembles something else in the index, it is scored and shown exactly like any other submission — and paid links are labelled as affiliate links wherever they appear.