AI face detection flags sections where the speaker is turned away, looking down, or has left the frame entirely. Get a complete log with timestamps before your edit reaches the client.
Viewers trust and engage with faces looking into camera. When a presenter glances away, checks notes or temporarily leaves frame, viewer attention drops and authority is undermined. Finding and cutting these moments manually in an hour-long interview is tedious, slow work. EditLint does it in under 60 seconds.
The subject's face is visible in frame but rotated significantly off-axis from the camera direction. This includes profiles, three-quarter turns and heads turned to look at off-camera prompts or other interview participants. Flagged when the rotation angle exceeds the configured threshold for the minimum duration.
No face is detected in the frame at all. This covers situations where the subject has stepped out of frame, ducked down momentarily, covered their face, or where the camera framing has drifted to an angle where the face detector can't find a face. The section is logged with start time, duration and whether audio was present.
The face is detected but the head pitch angle indicates the subject is looking down significantly — at notes, a phone, or a laptop. Less severe than turning away completely, but still worth flagging for longer durations. The minimum pitch angle and duration before flagging are both configurable.
The face visibility detector is designed specifically for talking-head video content — interviews, tutorials, vlogs, corporate presentations, podcast recordings and documentary subjects. It is not designed for coverage footage, drama or observational documentary where off-camera looks are intentional.
The detector produces a timestamped log of every section where face visibility drops below the configured threshold. Each item in the log includes the start time, duration, the reason it was flagged, whether the speaker was audible during that section, and a severity score.
The on-device face detector processes sampled frames at the configured rate (default every 6 frames). For each frame it returns all detected face bounding boxes, plus yaw (left-right rotation) and pitch (up-down tilt) angle estimates for the primary face. Processing is entirely local — no images leave your machine.
Consecutive frames are analysed for face presence and rotation angles. A rolling visibility score tracks whether the face is on-camera, on-axis and looking towards camera. Sections where the score falls below the threshold for the minimum duration are collected as candidate issues.
Each detected section appears in the EditLint panel with the reason, severity, duration and audio-presence flag. A colour-coded marker is placed on the timeline. You can click any item to jump the playhead to that moment and decide whether to cut the section, cover it with B-roll or leave it as-is.
Stop scrubbing through interview footage searching for off-camera glances. EditLint AI logs every face visibility issue in under 60 seconds.