Face landmarks and radial blur
![]()
Move your head and the center of a radial blur follows your nose across the camera image. The tracked point sets where the effect acts; the blur amount sets how strong it is.
Requirements and setup
Open the ZIP directly in score: it includes the RetinaFace model used by the patch. You need score’s ONNX/Pose Detector support. Despite the document’s filename, this model detects faces and five facial landmarks, not a full-body skeleton. To explore other models, use the LivePose model pack and select the appropriate Landmark Model. GPU inference also needs the acceleration libraries supported by your score build.
Select a working camera for Camera:/, grant camera access if prompted, and start playback facing the camera. Look for the annotated image and a nonzero detection count before adjusting the effect. The Object filter process requires the Jk data-query add-on.
Try it
Keep the blur amount fixed and move your head slowly from side to side. Then increase the amount: the blur center still follows the same path, but the effect becomes stronger.
The Object filter extracts the nose with .keypoints[2] and flips its vertical coordinate to match the effect. Change the index to another of the five landmarks to move the blur to an eye or a corner of the mouth. A separate plot follows the first landmark’s horizontal position; check it to confirm that motion is being detected.
If the camera image appears but the effect does not follow you, check the detection count and Landmark Model selection first. If movement is detected but appears inverted, inspect the coordinate conversion before changing the model.