How Does Our Facial Recognition AI Work?
In just a few seconds, Look Like Me tells you which parent you resemble most. But what's really happening under the hood? Here's how our LLM Engine analyses your faces.
Our LLM Engine: the technology behind the score
Look Like Me runs on its own proprietary AI engine — the Look Like Me LLM Engine, designed and trained specifically for family-based facial comparison. Unlike generic facial recognition APIs, it is optimised for a single use case: measuring resemblance between members of the same family.
As soon as you submit your three photos, the engine automatically detects faces and extracts what are called embedding vectors — a unique mathematical representation of each face, composed of hundreds of numerical values.
- Real-time face detection
- Extraction of key facial landmarks (eyes, nose, mouth, jaw, cheekbones…)
- Calculation of a high-dimensional vector for each face
This proprietary approach lets us own the accuracy, the privacy and the evolution of the algorithm — without depending on a third-party vendor.
Cosine distance: measuring similarity
Once the vectors are extracted, the algorithm calculates the cosine distance between your face and each of your parents'. The smaller the distance, the closer the vectors — and the more the faces resemble each other.
This calculation considers the overall geometry of the face rather than raw pixels. This means that two photos taken under different lighting conditions or at different ages still yield a reliable result.
The final score you see is a normalisation of this distance: 0% means no detectable resemblance, 100% means nearly identical faces (what you'd see comparing a photo of yourself to itself).
Test your family resemblance
Start a free analysis →What the AI doesn't do
It's important to understand the technology's limitations. Our LLM Engine analyses only facial geometry, not genes or DNA. It cannot predict who you'll look like in 10 years, nor identify potential genetic conditions.
It can also be influenced by external factors: heavy make-up, a thick beard, wide glasses, or poor image quality. For best results, we recommend recent, well-lit, front-facing photos.
Privacy and data deletion
Your photos are never shared with third parties. The facial vectors computed by our LLM Engine are not stored on our servers — only the final score is kept to display your result. Original photos are automatically deleted after 12 months of inactivity, in line with our GDPR policy.
You can request immediate deletion of your account and all your data from your profile at any time.
A constantly evolving technology
Our LLM Engine evolves continuously: we enrich the training datasets, refine the biases, and improve handling of photos taken under difficult conditions (low light, profile angle, large age differences).
The advantage of a proprietary engine: we can iterate on it as fast as we identify cases where it's wrong. Our goal is simple: give you a reliable answer to a question everyone asks themselves at some point.