Blur faces
detect faces with yunet + obscure them — wasi-nn on homelab
face-blur
blur is the default — produces the most natural-looking redaction. pixelate is harder to accidentally un-redact (looks like something was censored). black is the belt-and-suspenders option.
for blur, picks the gaussian sigma (1.0 at strength 1, ~40 at strength 100). for pixelate, picks the block size (4 px at strength 1, 40 px at strength 100). ignored for black.
lower = catches more borderline faces (risks blurring non-faces). higher = strict (risks missing faces). 0.60 is the opencv_zoo default.
download result
$ drag an image here
$ or click to browse
[ idle ]
input
output