FIELD GUIDE
Face masking in video: choose coverage, review motion, verify export
Hiding faces takes more than choosing a sticker. Check that the cover hides identifying features, follows the face during movement and remains present throughout the exported clip. This guide follows the full workflow and explains where automatic detection can fall short.
Confirm the tool matches the job
The detector applies one selected cover style and size to every face it detects. It is suitable when all detected faces should be covered. You cannot select only one person or draw an arbitrary mask area. If the goal is to hide one person while leaving others visible, confirm that this limitation fits before processing.
Prepare the source video and browser
Use a clip where faces are reasonably clear and visible. Small faces, low light, backlighting, motion blur and long profile views make detection harder. The tool reads MP4, MOV and WebM when the browser supports the file’s codecs. On first use, it downloads model and runtime files; the selected video itself stays in the browser and is not uploaded for processing.
Choose a solid cover or a transparent sticker
The solid option draws an opaque rectangle and makes it easy to judge whether facial features are hidden. There are also twenty portrait and animal head stickers. Their transparent outlines can leave parts of the original face visible, especially on a profile or when the face shape does not match the artwork. The tool does not offer blur or pixelation, so do not treat those effects as interchangeable.
Set the shared coverage size
The size control ranges from 60% to 200% of the automatically fitted face area; 100% is the default. Pause on a clear, near-frontal face and increase the setting until the recognizable features are covered. Preview the frame after each change instead of judging by the number alone.
The same size applies to every detected face. A value that covers a small, distant face may create an oversized block on someone closer to the camera. Find a workable balance across representative people and check each one. If no single size covers all faces without obscuring too much of the picture, this clip may not suit a shared global setting.
Too little coverage can leave facial edges visible; too much can obscure hair, clothing or nearby people. When concealment matters more than appearance, use an opaque solid cover and inspect its boundaries rather than choosing a sticker only because it looks more natural.
Preview key moments, not just the opening frame
Check when a person enters, a clear frontal view, a turn or downward glance, a brief obstruction, scene cuts and the end of the clip. New faces sometimes appear only near the end. For longer videos, pause at intervals as well as at visible action changes.
The current-frame preview answers whether the cover is aligned at that moment; it does not certify the whole clip. Compare the original and covered views at each sampled time. If no face is detected in that frame, move to where the face is clearly visible and check again.
Inspect tracking through difficult movement
Pay close attention to fast turns, people crossing, objects briefly blocking a face, and the frames around edits. A short detection gap may be bridged by tracking the recent position, but a face missed for longer can become visible again. Brief tracking prediction is not a guarantee that the cover remains accurate.
When a face turns sideways, its visible shape changes; enlarging the cover may help that person but also affects every other detected face. After a cut to a different person, confirm that the cover moves to the new face instead of lingering at the old position. Low light and motion blur can interrupt detection too.
Play each segment, pause at uncertain moments and step a few frames backward and forward. Check that the cover follows the face, disappears when the person leaves, and returns when they re-enter. If several frames are exposed, do not infer that later frames are safe from one successful still preview.
Process locally and understand the output
Detection, preview and export run in the browser. Keep the tab open during processing; closing or refreshing it clears the editing session. Processing often takes about as long as the video, depending on the device, browser, resolution and other running work.
The result keeps the original aspect ratio, is capped at about 1080p, and retains the source audio. The browser exports MP4 when recording support is available and may use WebM otherwise. The video is re-encoded, so check picture quality, playback compatibility and sound in the saved file rather than assuming it is bit-for-bit identical to the source.
Review the actual export from beginning to end
Play the downloaded file at normal speed once, then revisit the difficult points you noted earlier. Confirm that each face stays covered while present, that cuts do not expose a new face, and that movement does not pull the cover away. The final file—not only the preview—is the version to verify.
Check that a large cover does not hide a nearby person or important context, and that a small cover does not leave recognizable features at the edges. If you change a setting and export again, repeat the review; an earlier check does not apply to the new output. Keep the original separate so it is not mistaken for the processed copy.
Troubleshoot missed faces and failed exports
If a frame has no cover, check whether the preview detected a face there. Improve lighting or use a clearer source when possible. Increasing the global size may help a profile, but it changes coverage for everyone. If preview fails, try another time where the face is unobstructed.
If a file cannot be read, use a current desktop browser and a video with a supported picture track. H.264 video with AAC audio in an MP4 is often broadly compatible, though support still depends on the browser. For a slow or failed long export, close resource-heavy tabs and test a shorter prepared clip. The tool cannot manually draw a mask, isolate one person, or guarantee complete anonymity.
Use this sharing checklist
Before sharing, make sure each statement is true:
- The tool’s all-detected-faces behavior matches the people you intend to cover.
- The chosen style and size cover frontal, turned and moving faces.
- Entries, exits, brief obstructions and scene changes have been checked.
- The saved MP4 or WebM plays correctly and retains the expected sound.
- Names, captions, tattoos, license plates and other identifying details were reviewed separately.
- The file being shared is the reviewed export, not the unprocessed source.