We Built a Tool Just for Blurring License Plates. Turns Out That's a Bigger Need Than We Expected
We built a license plate blur tool thinking it was a tiny niche. 50K users later, I was wrong about who needed it.
I'm Yash Thakker, a bootstrapped founder. BGBlur is one of our products, and the team behind it is small: two ML engineers, two software engineers, two people in marketing, and me. No funding, no paid ads. BGBlur has crossed 50,000 users, almost all of it from organic search and word of mouth.
This post is about one feature that changed how I think about demand: a tool that does nothing except blur license plates.
Where we started
BGBlur began as a background blur tool. Creators wanted their face sharp and their messy room soft, without a green screen. Then we added face blur, because people kept asking how to hide someone in a video.
Technically, a license plate is just another object in the frame. You could already box it and blur it. So when someone on the team suggested a dedicated plate tool, my first reaction was "we already do that."
What changed my mind was our search data. People weren't searching for "object masking." They were typing "blur number plate in video" and "license plate blur" at 11pm with a dashcam clip open. They wanted one button, not a workflow.
That's the lesson I keep relearning: search terms describe jobs, not products. "Blur background" is a creative job. "Blur number plate" is a privacy job. Same engine underneath, completely different buyer, urgency and willingness to pay.
What it took to build
The detection part was the easy half. Our ML engineers had a working plate detector fairly quickly. The hard half was tracking.
A plate in a still photo is trivial. A plate on a car that changes lanes, gets hidden behind a truck for two seconds, and reappears at a different angle is not. If the blur slips for even three frames, the plate is readable and the whole export is useless.
Most of the ML time went there: keeping the blur locked on each plate as vehicles move and the camera shakes. The engineers built the pipeline around it so it runs on the web, on mobile browsers, and later through batch processing and an API.
Nobody ever told us "improve detection accuracy." They told us "it missed a plate at 0:42." Reading support messages closely gave us a better metric than any benchmark.
Who actually showed up
I pictured one user: the dashcam YouTuber trying to avoid an angry email. They came, along with motovloggers, whose footage is packed with plates, faces and street signs, all moving. Manual blurring is the kind of chore that makes people stop posting.
Then car dealerships started showing up. They shoot walkaround videos and listing photos constantly, with customer cars in the background. They needed it on photos as much as video, so we made sure JPG, PNG and WebP worked, not just clips.
Then fleets, delivery and parking operators. This is where "small niche" stopped being true. A fleet doesn't have one clip. It has hundreds of hours of footage. They didn't want a nicer editor, they wanted bulk processing and an API. A single-purpose page became the front door for our highest-volume work.
Journalists and public-sector teams came for faces and plates in the same pass, since street footage almost always has both.
The biggest surprise was AI and robotics teams collecting real-world video to train vision models. They capture thousands of plates and faces as a side effect and need to anonymize them before sharing or labeling. They aren't creators at all. They're data ops people with a compliance checklist.
Your first imagined user is often your smallest segment.
Why compliance changes everything
A license plate links back to a registered owner, so in many places it counts as personal data. The EDPB's Guidelines 3/2019 on video devices list licence plates alongside identifiable passers-by. Here in India, the DPDP Act is pushing companies to think the same way, and that's shaped a lot of our B2B conversations.
I'm not a lawyer, and rules vary by country. But the business insight is clear: once a problem carries even a little legal risk, "nice to have" becomes "must do before publishing." People buy faster.
Google figured this out long ago. Street View blurs faces and plates automatically, and their 2009 paper reported automatically blurring over 89% of faces and 94-96% of plates in their test sets. If you publish the road, you blur the road.
The blur that looks private but isn't
This surprised a lot of our users. A light Gaussian blur or coarse pixelation can look safe to a human and still leak information to software. In 2016, researchers from UT Austin and Cornell Tech showed off-the-shelf neural networks could identify faces and objects behind pixelation and YouTube-style blur, far better than people could.
That led to two product decisions. Strength and full coverage matter more than style, since a pretty blur that slips when a car turns protects nothing. And we added a solid black box redaction option, because for compliance teams "unrecoverable" beats "aesthetic" every time.
A mistake worth sharing
Growth hasn't been a straight line. At one point conversions dropped and we couldn't figure out why. It turned out to be hreflang and frontend errors on our locale pages, which were quietly breaking things for part of our international traffic. We traced it by lining up deployment history against our analytics and Stripe data. When your growth is organic, an SEO bug is basically an outage. We monitor it much more closely now.
What I'd tell other indie hackers
Ship single-job pages even when the engine is shared. Our plate tool shares most of its tech with the rest of BGBlur, but its own page, name and one-click flow is what made it discoverable.
Watch who shows up, not who you targeted. We aimed at creators and businesses walked in through the same door.
Put the pain in the headline. "AI video editing" means nothing. "Blur number plates in dashcam video automatically" is the sentence people already have in their heads.
Price for how the job arrives. Some people have one clip, once. Others have a thousand a month. That's why we went credit-based with small one-time packs and larger bundles instead of forcing a subscription.
A tool that looks too narrow is often exactly narrow enough.
If you're building in a "boring" privacy niche, I'd love to compare notes in the comments. And if you've got road footage lying around, try it at bgblur.com.
