How to blur every bystander but keep your subject sharp
Tell Vizard Agent which face must stay sharp and let it build a signature for that person. It tests the signature against strangers before blurring anything, runs the pass across every segment, then checks both failure directions: bystanders left sharp, and your own subject wrongly blurred.
What do you type into Vizard Agent?
Name the exception rather than the rule. "Blur the faces" is ambiguous about your own subject. The version that works tells Vizard Agent who stays sharp and treats every other face as the default case. Reference faces from different angles beat one perfect portrait, because recognition has to work on profile and half-turned shots too.
Prompt
What does Vizard Agent actually do?
On a nineteen-minute vlog filmed in a busy street, the key step came before any blurring: testing the signature against strangers to see who wrongly matched. Afterwards both failure directions were debugged, because a pass that only checks for missed strangers can blur the presenter for four seconds and nobody notices. In the session behind this article it:
- Built the subject's face signature, confirmed the reference crops were all the same person and tested recognition against strangers' faces.
- Tuned detection to catch profile faces, which the default misses, and checked the subject was recognised across the whole video.
- Tested the blur pass on two tricky segments first, expanded the references and re-checked that strangers stayed unmatched.
- Ran the blur across all fifty-eight segments and spot-checked it at key moments.
- Used face-track scores and track-level similarity to debug a segment where a bystander stayed sharp, re-ran with refined identity rules, then fixed another where the subject was still blurred.
- Checked whether the source already had blur baked in, and extended the blur to a person who appeared again later.
When does this not work well?
Recognition needs faces it can actually resolve in the frame. Distant crowds, heavy motion blur and low light reduce Vizard Agent to guessing, and a blur pass that guesses is not a compliance measure. Masks or helmets can make detection skip people, and twins or close relatives may both match the signature. This is a tool, not legal advice.
Common questions
Will it blur me by mistake? It can, which is why Vizard Agent tests for that specifically before delivering.
What about profile shots? The default detection misses them. Vizard Agent tunes for them deliberately.
How many reference faces do I need? A handful from different angles. Vizard Agent will ask for more if the matching is shaky.