Sigmora

Sigmora Safe Cut: publish real-world footage without exposing the people who never agreed to be in it. Creators who film in public (street interviews, vlogs, cafes, events) publish strangers, children, number plates and screens full of personal data every day. Today's tools either blur every face, including the guest, or leave hours of masking by hand, so most creators just publish. Safe Cut decides from context what shouldn't be published, and proves it: 1. SAM 3.1 tracks every face, person, plate, screen and document through the clip. 2. Muse Voice Transcribe hears who speaks. 3. Muse Spark watches the whole clip, then rules three times, independently, on numbered keyframes, about who is part of the story. Disputes escalate once at maximum effort. A miss loop asks what the tracker didn't catch. 4. A deny-by-default policy hides everything uncertain. Children are always hidden. A person the tracker loses and finds again comes back hidden. 5. It hides identity, not the shot: every cover is one soft frosted look (a coarse locked-grid mosaic under a feathered blur, so it can't be reversed), for faces, plates, screens and documents alike. A kept participant's face is never covered. 6. The creator reviews every item, sees why each was decided (keyframe, three votes, transcript line), and approves. 7. Before anything is published, an independent face detector checks every rendered frame and a Muse Spark red team checks sampled frames. Leaks are patched automatically. 8. A hash-chained privacy receipt records every decision, who made it, and how the data was handled. Everything is deleted after 24 hours. It's a production service: an API on Modal (scales to zero) with access codes, per-clip and daily spend caps, signed uploads and 24-hour deletion; a review web app; and a CLI. It's built to plug into Sigmora, our creator platform. Try it: https://safecut.sigmora.org → Try it → access code qMKU1rfUqTkzbLRI8QzD (valid to Nov 10, daily spend cap). The demo footage is AI-generated: no real people.