DynsimulationTechnologies
Automated Face Anonymization & Privacy Redaction API

Face Blur & Privacy Redaction
Automated Multi-Face Anonymization in 90ms

An identifiable face turns an ordinary photo into special-category biometric data. This API finds every face in a photo, an archive or a live RTSP stream and masks it before the frame is ever stored — so what you keep is analytics, not personal data.

Try it below with your own image, your webcam, or one of the sample scenes.

Try Live Interactive DemoRequest API Access
Recall-First
Tuned to Miss No Face
<90ms
Per-Frame Latency
GDPR Art 9
Biometric Data Scope
RTSP & Batch
Video Feed Ready
Live Demo API

Live Face Blur & Redaction Sandbox

Upload any photo, capture from webcam, or pick a guaranteed sample preset to test automatic face redaction.

Presets:
Source Image for Redaction
Source Image for Redaction
Samples:
Redaction Parameters

Configure Anonymization Mode

Blur Radius Intensity25px
Interpretation Guide

Choosing the Right Redaction Mode

The three modes differ in how much of the original scene survives the mask. Weak settings on any of them can leave a recoverable signal, so the radius matters as much as the mode you pick.

ModeRecoverabilityVisual EffectBest Suited To
Gaussian BlurIrreversible above radius 15Scene stays readablePublished photography, press imagery and marketing footage where the crowd should still look like a crowd.
PixelateIrreversible at block size 12+Obvious visual signalBroadcast and social media, where viewers need to see that redaction was deliberately applied.
Solid BlackoutFully irreversiblePixels discarded entirelyEvidence handling, legal disclosure and regulator submissions, where no residual signal may remain.

A light blur is a cosmetic effect, not a redaction — low-radius blur and coarse pixelation can both be partially inverted. If the output is evidential or leaves your control, use solid blackout.

How It Works

Four Stages, About 90ms per Frame

The whole path from ingest to anonymized output runs in memory. There is no queue to poll and no temporary file to clean up afterwards.

01

Ingest

Accepts JPEG, PNG and WebP uploads, batch archives, or a live RTSP/RTMP stream. Frames are decoded straight into memory — nothing is written to disk at any stage.

02

Detect

A single-pass detector locates every face in the frame, including profiles, partial occlusions and faces as small as 20×20 pixels in dense crowd scenes.

03

Redact

Elliptical masks are fitted to each detection and filled using your chosen mode and radius. The mask follows the face contour, so background detail outside it is left untouched.

04

Return

You get back the anonymized buffer plus telemetry — face count, bounding boxes and processing time — then every intermediate buffer is discarded from memory.

Developer First

Integrate Automated Face Blur

Call our REST endpoint with image files or RTSP video frames. Receive anonymized image buffers with zero personal data stored on disk.

  • Single-pass multi-face detection & Gaussian elliptical masking
  • Configurable blur radius, pixelation, or solid blackout
  • Zero media storage retention — strictly ephemeral memory processing
curl -X POST https://api.dynsimulation.com/v1/vision/face-blur \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "image=@crowd_photo.jpg" \
  -F "mode=gaussian_blur" \
  -F "radius=25"
Where It Is Used

Media You Need to Keep, Without the Identities

The pattern is the same across industries: the footage has genuine operational value, but the faces in it carry obligations you would rather not take on.

CCTV & Retail Analytics

Keep footfall and dwell-time analytics running while faces are stripped from every stored frame, so the recording itself never becomes biometric data.

Dashcam & Mobility Fleets

Redact bystanders and other drivers before telematics footage leaves the vehicle, which is what makes road-scene datasets shareable with partners.

Healthcare & Clinical Media

Anonymize ward photography, procedure recordings and patient intake media so that clinical records can be used for training without exposing identity.

News & Publishing

Protect minors, protest participants and uninvolved bystanders in published imagery, with an audit record of exactly how many faces were masked.

Insurance & Claims

Sanitize accident and site-inspection photos before they are passed to adjusters, third-party assessors or subrogation partners.

Smart City & Public Sector

Run crowd density, queue and traffic monitoring on public feeds while satisfying the requirement that no identifiable individual is retained.

Deployment

Run It Where Your Media Already Lives

Same detection model in all three; the difference is where the frames are processed and who operates the runtime.

Managed API

Fastest path to production

  • Single REST endpoint, no infrastructure to run
  • Autoscaling with regional processing endpoints
  • Usage-based pricing with a sandbox tier

RTSP Stream Connector

For live camera estates

  • Subscribes to existing RTSP/RTMP camera feeds
  • Redacts in-line, before frames reach your recorder
  • Re-publishes a clean stream to your VMS

On-Premises Container

For restricted networks

  • Docker or Kubernetes image inside your perimeter
  • Runs fully air-gapped with no outbound calls
  • Per-node licensing with offline model updates
Compliance & Data Handling

Redaction at Ingest Beats Deletion Later

Once identifiable faces land in your storage, they carry retention limits, access controls, subject-access obligations and breach exposure for as long as you hold them. Masking before the write removes the obligation instead of managing it.

Talk to our compliance engineers

GDPR Article 9

A face in a stored image is special-category biometric data. Redacting at ingest means the data you retain falls outside that category entirely, rather than needing a lawful basis to hold it.

HIPAA Safe Harbor

Full-face photographic images are one of the eighteen identifiers Safe Harbor requires you to remove before media can be treated as de-identified.

Zero Retention

Media is processed strictly in memory. No original frame, intermediate buffer or detection crop is ever persisted to disk, and nothing is used for model training.

Audit Telemetry

Every response carries the face count, bounding boxes and processing time, giving you a per-asset record that redaction actually ran and what it covered.

This page describes product capabilities and is not legal advice. Your own counsel should confirm how redaction fits your lawful basis and retention policy.

Deploy Face Blur

Request Face Blur API & Video Stream Credentials

Connect with our technical team to receive sandbox API keys, RTSP video pipeline connectors, or on-premises container licenses.

Frequently asked questions

What redaction modes are available?

Elliptical Gaussian blur, pixelation and solid blackout, with a configurable radius so the strength of the redaction can match your compliance requirement.

Does it work on video as well as photos?

Yes. It handles crowd photos and CCTV stills in batch, and redacts live RTSP video streams through a continuous pipeline.

How accurate and how fast is detection?

A stated 99.6% face detection recall at around 90ms per frame.

Which regulations does this help with?

It is built for GDPR Article 9 obligations around biometric data and for HIPAA, by removing identifiable faces from media before it is stored, shared or published.

Is any media retained?

No. Media is processed strictly in memory with zero storage retention, so nothing is written to disk during redaction.

Can it be deployed on our own hardware?

Yes. It is available as a managed API, with RTSP pipeline connectors and on-premises container licences for deployments that must stay inside your network.

Explore the rest of the suite

View all products