In short: face tracking finds key points on your face, such as your eyes, nose, mouth and jawline, many times a second, then glues digital effects to those points so they move and rotate exactly as you move. Flip the switch to see the points, then the effect.
The dots are the tracked points. The dog ears and nose are pinned to them, so they tilt and shift exactly as the face does.
The camera doesn't need to know your face. It only needs to keep finding the same handful of regions, frame after frame.
Curious about the full mechanism? Read how a Snapchat Lens gets built.
Same landmark map, three very different results depending on what gets attached to it.
Official Snap partner studio, 6 years directing AR campaigns. Replies within 48 hours.
Quick answers
What is face tracking?
Face tracking is a camera system that finds key points on your face, such as your eyes, nose, mouth and jawline, many times a second. Digital effects are glued to those points, so they move and rotate exactly the way your head and expressions do.
How many points does face tracking use?
Most face filters track somewhere around 70 or more points across your face, covering the eyes, eyebrows, nose, mouth and jawline. That map is rebuilt dozens of times every second, so effects stay locked on even while you talk or turn your head.
Does face tracking work on more than one face at once?
Yes. Most modern face tracking systems, including Snapchat's, can find and track several faces in the same camera frame at the same time, applying the same or different effects to each one.
Why does a filter sometimes flicker or fall off my face?
Face tracking needs a clear, well lit view of your face. It can lose lock if your face turns too far to the side, gets partly covered, or the lighting is too dark, because the system can no longer find enough of the key points to place the effect confidently.
Is face tracking the same as face recognition?
No. Face tracking only maps the shape and movement of a face to place effects on it. It does not identify who the face belongs to. Face recognition is a separate technology built to match a face against a known identity.