Beyond Faces: The Future of License-Plate, Body, and Object Anonymization

A face is not always the thing that needs hiding. A delivery van's plate, a patient's wristband, a distinctive uniform, or a name on a screen can matter just as much.

Unseen currently focuses on faces. License-plate detection, whole-body anonymization, custom-object tracking, segmentation, and object removal are possible future directions—not features you should expect to find in the current editor.

If one of those objects must be hidden in today's export, plan a separate edit.

A moving box and an object outline solve different problems

A rectangular mask is often enough for a name on a fixed screen. When the camera moves, the mask has to follow the region. When the screen turns away or leaves the frame, its timing needs adjustment too.

Segmentation tries to follow an object's shape more closely. That could make sense for hiding a whole person while leaving nearby equipment visible. It also creates a harder review problem around fingers, hair, thin straps, and objects crossing in front.

Meta's Segment Anything Model 2 is one example of technology for prompted object segmentation in images and video. It is relevant to this direction, but it is not currently integrated into Unseen. A research model's capabilities do not establish how a future product feature would behave on your recording.

Masking is not the same as removing an object

Covering a plate with an opaque region makes the edit obvious. Removing the plate and filling its place with plausible imagery changes the scene.

That distinction matters when the recording is meant to document what happened. A conspicuous redaction can communicate that information was withheld. A reconstructed background may look like an untouched image even though it contains generated or inferred content.

Before choosing an effect, decide whether your goal is to conceal information or create a different-looking scene.

What to do with mixed footage now

Suppose a security clip contains a visitor's face, a vehicle plate, and an address on a package.

Use face masking for the face if it suits the task. Handle the plate and address in an editor that supports manual masks or object tracking. Alternatively, crop them out or use a shorter section where they are not visible.

Review the combined final export after the last editor has touched it. Check the plate as the vehicle turns, the package as it moves between hands, and the first and last visible frames of each object. A face list is not an inventory of every sensitive thing in the scene.

What a useful future workflow would need

A strong object-redaction feature would need more than a “select object” button. It would need understandable selection, tracking across obstructions, visible failure points, correction tools, and reliable export review.

Those are evaluation criteria, not a delivery promise. Until such a workflow exists, avoid designing a project around an assumed plate detector or SAM2 integration. Choose tools based on the controls you can use and test today.