Face Recognition Service
An original, in-house face-recognition engine that searches millions of faces in about a second flat — built to power DIVO Global's paid Face Search feature for finding look-alike models.
DIVO Global's flagship feature needed face recognition with high accuracy on faces, emotion, gender and ethnicity, able to find look-alikes at scale.
The service had to hold up under high load with millions of stored faces.
Built an original, in-house recognition engine — not a wrapper around a single off-the-shelf model — combining InsightFace, OpenCV and our own trained models into one pipeline tuned specifically for fashion and portrait photography.
Engineered the similarity index and vector search so a query returns ranked matches from millions of stored faces in roughly one second, at production load.
Built an admin and management app plus an analytics dashboard for usage and requests, and trained the model to read gender, race, age and emotion reliably across a global, highly varied user base.
The hard engineering behind the result.
Reaching sub-second search across millions of faces without sacrificing accuracy.
Handling the diversity of real fashion photography rather than lab-quality images.
Training the model to read gender, race, age and emotion reliably across a global, highly varied user base.
Face Search, fully integrated into the product.
A look at what we shipped.
A happy customer.
The feature became DIVO's "killer feature," helping it stand out from competitors and attract up to $1M in investment; Saturn Labs continues to support and extend it. The same engine now ships as the Face Search module inside the upcoming DIVO Global relaunch — see it in action in the integration video above, or try the technology yourself at the live preview.