Automatic face masking for video

Do I have to mask a 12-minute video frame by frame?

No. Upload the video once and call encrypt_video. YOLO-based detection runs on every frame automatically, so you never scrub through the timeline painting mosaics by hand. Long media is processed as a background job that you can poll and download when it finishes.

Can it blur just the faces, without picking coordinates?

Yes. The server detects faces first and blurs only the detected face regions — it is not a full-image blur filter. You never click boxes or type coordinates. Ask for target_objects=["face"] for face-only masking, "body" to mask whole people, or "car" for plates and vehicles.

Several people appear at once. Can it mask them all in one pass?

Yes. Batch face detection and video masking happen in the same call. Every person detected in each frame is covered, including people walking through the shot — the mask follows them because detection re-runs on every frame.

What if I need the original later?

Masking here is encryption-based, not a destructive paint-over. With encryption keys, an authorized account can restore the original or partially decrypt only selected objects. A plain mosaic smeared over pixels cannot be undone — this can, for viewers you authorize.

No. It is not video inpainting or object removal: people are covered on every frame, not erased from the footage. If you were searching for how to remove a person from a video automatically, masking is the evidence-safe alternative — the recording itself stays intact for authorized review. That is what makes it suitable for video de-identification: privacy compliance without destroying the source.

encrypt_video(target_objects=["face"])
encrypt_image(target_objects=["face"])

• Single call: face detection plus face blur/redaction, with no separate detect tool + blur tool chain.
• Detection runs on every frame — people moving through the shot stay covered.
• Only detected regions are masked, never the whole frame.
• Long media runs as a pollable job with a download when it completes.
• Remote MCP server over Streamable HTTP with OAuth2 sign-in — works with standard MCP clients.