Klaimed's candles vision model understands a candle like a category expert — vessel or pillar shape, wick type and wax surface, lid, label and finish — to find the exact product from a reference and its closest matches. The vision foundation for brand protection.
The model reads a candle the way a specialist in the category would — vessel or pillar shape, wick type and count, wax colour and surface, lid or closure, label placement, finish — resolving the shape, geometry, material, colour and type of each, and how they combine into one design.
Reads candles the way a category expert would: overall shape and proportion, the geometry of each component, material and surface character, and the specific construction features that define one design against another. The output is a structured record of what is actually visible, not a single label.
Given a reference product, returns the closest matches from a monitored set, ranked by design similarity rather than by colour, scene or filename. Two candles that share construction sit together even in different colourways; two that merely share a palette do not.
Every match comes with its reasoning: which features agree between the reference and the candidate, which diverge, and how confident each comparison is. A reviewer sees why a pair was matched instead of an opaque similarity score.
The same design retrieves consistently whether it was shot as a studio packshot, a seller's phone photo or an in-use image, and holds up under changes of angle, lighting, background and partial occlusion.
Every model is trained against a structured attribute taxonomy: the named visual features that separate one design in this category from another, and the values each of those features can take. The taxonomy is proprietary and the training dataset as well.
We will talk you through what the model is trained to look at for your category, and if a feature matters to your workflow that we have not covered yet, we add it to the taxonomy and retrain. The taxonomy grows with the categories our customers actually work in.
Tell us the category and the volume. We will show you the model running against your own listings before anything is signed.