How do Face Search Tools Work? Facial Recognition AI Models Explained.

What Is a Face Search Tool?

You probably know face search. It’s a system that finds people on the internet based on their photos.

In simple terms: you upload a photo of yourself, and face search tools find you on other websites.

How Do You Build an AI Model?

Let’s start with hardware - For bigger AI models, developers usually use computers with multiple powerful GPUs. GPUs can do many mathematical calculations at the same time.

Then, developers write code. Python and AI frameworks such as PyTorch or TensorFlow are commonly used for defining what the model should learn and what data is used for training.

Once the code is ready, they collect images and train the model. The model is shown huge numbers of examples. At first, its answers are mostly wrong. Each time it makes a mistake, its internal settings - called weights - are adjusted slightly, until the results are satisfactory.

Below, we will explain how AI face search models find and compare faces in images, and how images are found online.

How Does Facial Recognition Work?

Facial recognition works in a few steps. Here’s the simplest explanation:

1. Recognizing a face in a photo

The first step is finding the face in the uploaded image. The system detects where the face is and separates it from the background and other objects in the photo.

How are Facial Recognition Models taught?

Facial Recognition models are trained using large numbers of face images. During training, the AI learns which facial features are useful for telling people apart. Over time, the model learns to recognize similarities between different photos of the same person.

Simply said: you “feed” AI so many images that it learns to tell them apart.

2. Facial characteristics analysis

After the first step, AI analyzes the face. It finds the distances between each feature. Then, it returns a mathematical representation of its findings.

3. Converting the face into a vector

Instead of storing information such as "large eyes", the system represents these characteristics using numbers. This makes it much easier and faster for computers to compare one face with millions of others.

4. Comparison

AI compares the facial features it detected with the faces found in the index or a database. Faces with similar vectors are considered possible matches.

The system can then calculate a similarity score and show the closest results first.

Crawlers - how images get collected

For a face search engine to find images, it first needs to build an image index. This is done using web crawlers.

Crawlers are automated programs that visit publicly accessible webpages. When an image is found on such a page, the search engine adds it to its index. If a face is detected, the system can create a facial vector that can later be compared with uploaded photos.

Facial Search Models vs LLMs - Is It the Same Thing?

There is some confusion regarding face search and LLMs, so here’s a simple answer.

No, facial search models and LLMs are not the same thing, although both are types of AI models.

LLMs are models that generate text - such as Chat GPT or Google Gemini.

Therefore, face search tools like lenso.ai or eyematch.ai are not LLMs, but rather reverse image search engines that use their own AI Models.

Lenso.ai is also a non-generative AI - not an LLM. You can learn more about the differences between generative and non-generative AI in this article: Generative AI vs. non-generative AI - what are the differences?