LensBridge AI Fixes Blurry, Glare-Washed Photos From Cheap Cameras

Every camera lens is a little bit imperfect. Cheap, tiny lenses — the kind squeezed into phones, doorbell cameras, drones and medical scopes — blur and smear images in ways that get worse in bright light. The worst offender is "veiling glare": a milky, washed-out haze that settles over a photo, like looking through a smudged window on a sunny day. Software can fix some of this, but only if someone has already fed it matched "bad photo, good photo" pairs from that exact camera, which is expensive and often impossible to gather.

LensBridge, from researchers in China, takes a two-step approach. Step one builds a general-purpose repair model trained on a large library of known lens flaws, so it can recognize and undo blur without being told which lens it's looking at. Step two adapts that model to glare using only a few ordinary photos from the target camera — no paired examples required. The trick is splitting each image into broad, hazy shapes (glare) and fine, crisp detail (the subject), then training separate branches to handle each.

The "no paired examples" detail is the real breakthrough. Normally, teaching AI to clean up a specific camera means shooting the same scene twice, with and without the problem — practically impossible for glare, which depends on where the light is. By working from unpaired snapshots, LensBridge could be pointed at almost any camera, including ones already in the field.

In tests across multiple optical systems, the method successfully removed blur and glare together, something earlier tools handled only one problem at a time. The authors say they'll release the code publicly, which means camera makers could eventually build it in. The honest caveat: this is an academic paper, not an app you can download today, and results were verified on specific test setups — not your phone. But it points toward a future where cheap hardware takes expensive-looking photos.