Attributions
What this is built from
The detection models are other people’s work. This page names each one, its licence and where it came from, along with the open-source libraries doing the decoding, inference and muxing.
Free with a small watermark · £10 one-off to remove it · your video never leaves your browser
Detection models
- UltraFace (MIT) — the two fast face detectors, “Faces — fast” and “Faces — fastest”. From Linzaer/Ultra-Light-Fast-Generic-Face-Detector-1MB, served via jsDelivr.
- YOLOv11m-face (MIT) — the “Faces — high quality” detector. From astaileyyoung/yolov11m-face-onnx on Hugging Face.
- YOLOv11 licence plate detection (AGPL-3.0) — the “License plates” detector. From morsetechlab/yolov11-license-plate-detection on Hugging Face. It is a finetune of Ultralytics YOLO11, which is itself AGPL-3.0.
Where the models run
All three are downloaded to your browser and run there. They are not hosted or executed on any server of ours, and no video, frame or detection is sent anywhere for them to work — the same property described on verify it yourself, which lists the model hosts among the requests this site makes.
The two Hugging Face models are fetched from Hugging Face directly, which redirects the download to its own regional CDN. Those hosts see that a model was requested and your IP address, and nothing else.
Libraries
- ONNX Runtime Web (MIT) — runs the detection models, on WebGPU where available and multi-threaded WebAssembly otherwise.
- mp4box.js (BSD-3-Clause) — parses the container and extracts encoded samples.
- mp4-muxer (MIT) — writes the redacted frames and passed-through audio back into an MP4.
- Svelte (MIT) — the interface.
- Melt UI (MIT) and Radix Colors (MIT) — accessible interface primitives and the colour palette.
- Stripe.js (MIT) — only on the watermark-removal purchase.
Corrections
If something here is wrong, out of date, or your work is used without proper credit, tell us and it will be fixed: hello@anonymizevideo.app.