The selfie risk score (the value shown beneath a user's selfie) is separate from all the facial attributes Plaid checks (nose, eyes, etc.) and takes into account things like image clarity, liveness detection (e.g. picture of a picture vs. real person), number of frames captured, and other risk signals from the image (e.g. indicating a deepfake or the user is wearing a mask). If users score medium or higher for the selfie risk score, they’ll fail the selfie check — even if our algorithms were able to verify that the selfie matches the photo extracted from their ID.