AnonymizeVideo.app
New Member
I've built a tool for blurring number plates and faces in dashcam footage and I'm looking for people to test it properly before I promote it anywhere wider.
The short version: anonymizevideo.app does the whole job locally in your browser. Nothing is uploaded. It uses WebCodecs to decode and re-encode via your machine's own media hardware, and runs the detection model through ONNX Runtime on WebAssembly, with WebGPU acceleration where the browser supports it. After the first visit it's cached, so you can pull your network connection and it still works — which is the easiest way to satisfy yourself that nothing is being sent anywhere.
Workflow is: load an MP4 or MOV, pick the face model or the licence plate model, run the scan, review the detections, export.
Technical details people here will ask about:
- Input: MP4 and MOV. Export: H.264/AVC, VP9 or AV1, with control over output resolution and target bitrate, and the option to preserve the original audio track
- Length: about 15 minutes at 1080p on typical hardware. The finished file is assembled in RAM before download, so the ceiling is your own memory rather than any server quota. Scaling the output resolution down makes long jobs far more likely to complete
- Browsers: anything with WebCodecs — Chrome, Edge, Safari 17+, recent Firefox. Desktop is considerably more dependable than a phone for anything beyond a short clip
- Detections are tracked across frames, so redactions stay attached through brief occlusions and changes of angle
- Per-detection control of blur, pixelate or solid block, adjustable confidence threshold, adjustable region size and edge softness
Free, no account required. There's a small watermark on free exports and a £10 one-off payment removes it — mentioning that here rather than burying it.
What would be genuinely useful: results from cams and conditions I can't easily test. Non-UK plate formats, older or unusually shaped plates, night and low-light footage, rain and windscreen glare, 60fps and 1440p/4K files, and anything from cams that write in less common containers. If detection fails I'd like to know the cam model, the settings, and roughly what the conditions were.
Happy to answer anything about how it works under the bonnet.
The short version: anonymizevideo.app does the whole job locally in your browser. Nothing is uploaded. It uses WebCodecs to decode and re-encode via your machine's own media hardware, and runs the detection model through ONNX Runtime on WebAssembly, with WebGPU acceleration where the browser supports it. After the first visit it's cached, so you can pull your network connection and it still works — which is the easiest way to satisfy yourself that nothing is being sent anywhere.
Workflow is: load an MP4 or MOV, pick the face model or the licence plate model, run the scan, review the detections, export.
Technical details people here will ask about:
- Input: MP4 and MOV. Export: H.264/AVC, VP9 or AV1, with control over output resolution and target bitrate, and the option to preserve the original audio track
- Length: about 15 minutes at 1080p on typical hardware. The finished file is assembled in RAM before download, so the ceiling is your own memory rather than any server quota. Scaling the output resolution down makes long jobs far more likely to complete
- Browsers: anything with WebCodecs — Chrome, Edge, Safari 17+, recent Firefox. Desktop is considerably more dependable than a phone for anything beyond a short clip
- Detections are tracked across frames, so redactions stay attached through brief occlusions and changes of angle
- Per-detection control of blur, pixelate or solid block, adjustable confidence threshold, adjustable region size and edge softness
Free, no account required. There's a small watermark on free exports and a £10 one-off payment removes it — mentioning that here rather than burying it.
What would be genuinely useful: results from cams and conditions I can't easily test. Non-UK plate formats, older or unusually shaped plates, night and low-light footage, rain and windscreen glare, 60fps and 1440p/4K files, and anything from cams that write in less common containers. If detection fails I'd like to know the cam model, the settings, and roughly what the conditions were.
Happy to answer anything about how it works under the bonnet.