Polygon-precise face and licence-plate blur. Single-process pipeline for speed, H.264 and H.265 input, H.264 output, SHA-256 integrity hash. GDPR and FOIA compliant by default.
Manual frame-by-frame redaction costs more than most investigations are worth. Automation changes the calculus.
GDPR Article 17 ("right to be forgotten") and FOIA redaction requirements both require that identifiable third parties be masked before release. Manual frame-by-frame redaction is prohibitively expensive, Visylix automates it with court-ready integrity hashes.
Security teams routinely share footage with law enforcement, insurance adjusters, and legal counsel. Automatic redaction of non-subject faces and plates removes the privacy-exposure risk from every export, no more accidentally publishing a bystander's identity.
Visylix redaction is baked into the compressed bitstream. There is no reverse transform, no "unblur" button. The SHA-256 hash on the output file proves nobody modified the redaction after the fact, critical for chain-of-custody in regulated industries.
No subprocess fork-bombs. No intermediate MP4 writes. Just a tight native pipeline.
Face detection finds every face in frame and tracks them across motion, then applies a colour-true polygon-precise blur. Only facial pixels are redacted, the surrounding scene is preserved.
ANPR detects plates, including partial and angled views. Polygon follows the plate quad frame-to-frame. Ideal for public release of traffic footage or sharing evidence across jurisdictions.
Ray-casting point-in-polygon test means the blur follows the exact detected region, not a loose rectangle. Head rotation tracked. Plate quads tracked. No accidental over-blur of surrounding evidence.
Decode → detect → blur → encode runs in one tightly integrated process. No external transcoder subprocess, no Python video bridge. A 10-minute clip redacts in roughly 90 seconds on a typical VMS host.
Every redacted output file is hashed with SHA-256 at write time. The hash is stored with the clip metadata and included in exported bundles. Tamper-evident by construction.
POST a job, poll progress, cancel if needed. Queue dozens of jobs for long investigations. Engine parallelizes across available GPU / CPU. Partial output discarded on cancel.
Four stages, one process, zero external tools.
H.264 or H.265 input decoded in-process. No re-muxing, no intermediate file. Direct frame access for the detectors.
Face detection and/or licence-plate detection run inline. Bounding regions produced per frame. A tracker maintains identity continuity across motion.
Polygon-precise colour-true blur over the detected region, preserves evidence around the face or plate while fully masking the identifier.
The modified frames are re-encoded to H.264 MP4. SHA-256 is computed incrementally. Output is written together with its integrity hash.
Regional hospital, 340 cameras. HIPAA-ready workflow: clip selection → auto-redact non-subject faces → SHA-256 integrity bundle → sealed release to patient counsel.
“Right to be forgotten” mandates removal of identifiable PII. Visylix redaction provides irreversible face masking with cryptographic proof.
Public-records release of surveillance footage requires non-subject masking. Automate the highest-cost step in FOIA response.
De-identification of protected health information. Visylix face redaction aligns with the Safe Harbor method for video PHI.
SHA-256 integrity hash on every output file. Tamper-evident. Pair with Evidence Lock for full court-admissibility workflow.
Scale includes face redaction. Add ANPR for plates. Enterprise bundles both , plus polygon editing for arbitrary regions.
You select a clip and pick a redaction preset, faces, license plates, or both. The Visylix engine decodes the clip (H.264 or H.265 input), runs the relevant detector (Face AI for faces, ANPR for plates), applies a colour-true polygon-precise blur to each detected region, then re-encodes to H.264 MP4 with a SHA-256 integrity hash. The whole pipeline runs in a single process, no external transcoder subprocess, no Python video bridge, so a 10-minute clip redacts in roughly 90 seconds on a typical VMS host.
Detections return bounding boxes from the AI model, then a polygon-precise clip is applied using ray-casting point-in-polygon tests. This means only pixels inside the detected region are blurred, background context is preserved. For faces, the polygon tracks head rotation frame-to-frame. For plates, the polygon follows the plate's bounding quad. No rectangular over-blur, no accidentally covered evidence.
No. The redaction re-encodes the video with blurred regions baked into the compressed bitstream. There is no reverse transform. This is intentional, redaction is only valid for compliance and evidence sharing if it cannot be undone. The SHA-256 hash on the output file provides tamper-evidence: if a single byte changes after redaction, the hash changes, and downstream systems can detect it.
Yes. Every redaction is an async job queued in the engine. You POST a job, get a job ID, and poll the status endpoint for progress percentage and estimated time remaining. Cancel at any time, partially-rendered output is discarded. For long investigations, you can queue dozens of jobs and the engine chews through them in parallel based on available GPU / CPU resources.
Face redaction is included on Scale and Enterprise (requires Face AI). Plate redaction is included on Scale and Enterprise with the ANPR add-on (requires ANPR). Enterprise tier gets both face and plate redaction together, plus polygon editing tools for manual redaction of arbitrary regions (useful for exposed whiteboards, screens, or sensitive documents captured in frame).