The moment I understood why similar photos matter more than you think
A customer emailed last month with a simple complaint: she'd taken 47 photos of her nephew at a birthday party. Same angle, same five-second window, slightly different expressions. Her camera roll had become a museum of near-identical shots, and she had no system for picking the keeper. That email changed how I thought about photo clutter.
Why your camera roll is full of 'close enough' photos
If you shoot on your phone with any regularity, you know the problem. You see a moment worth capturing. You hold the shutter button for three seconds to burst-shoot it. You get six versions of almost the same frame. Maybe one has better light, one has better focus, one has someone blinked. But your camera roll treats them as six separate events.
That's not a problem unique to amateurs. Event photographers, wedding shooters, anyone who culls thousands of photos - they deal with the same friction. You end up with phantom duplicates that aren't quite duplicates. Your eyes can spot the difference between them, but it takes time. A lot of time.
We heard this enough times in early user feedback that it became clear: a feature that could group similar photos wouldn't just clean up storage. It would save time during the actual culling workflow.
How we actually detect 'similar' without getting it wrong
The technical challenge sat in the middle of two extremes. We could flag obvious duplicates - same file, different metadata, simple hash comparison. But 'similar' is hazier. You might have three shots of the same moment, each slightly cropped or edited. Or you might have two completely different photos taken from the same location, same lighting, same day. One is a duplicate to cull. The other is a keeper shot you want to compare side by side.
We use Apple's Vision framework to build a feature print - a mathematical signature of each photo that captures visual content. Two photos with similar content get grouped together. But here's the part that matters: we don't auto-delete anything. The grouping surfaces candidates. You still decide which frame is the keeper.
The algorithm isn't perfect, and we knew it wouldn't be. But it's efficient enough that culling time dropped measurably in our internal testing. And because it's a Plus feature (not free), we're being transparent about its scope: it works best on burst sequences, event shooting, and handheld multi-shots. It's not a magic wand. It's a lens.
The real story: why we didn't ship this in week one
Similar-photo grouping took us longer to build than we expected, and we made a deliberate choice to ship it late rather than ship it broken. In our first pass, the clustering was too aggressive. We'd group a photo of your friend at a café with a photo of a different friend at a different café because the Vision signatures caught 'people + warm indoor lighting' and treated them as matches.
We spent weeks tuning the threshold. We ran it against real camera rolls from beta testers - wedding photographers, parents, travel snappers, people with 10,000+ photos. We watched it work and fail, adjusted the sensitivity, watched it work differently. We did not ship it until the false-positive rate was low enough that users felt like the grouping was useful, not noisy.
That's why similar-photo grouping landed as a Plus feature, not free. It required enough engineering and validation that it made sense to reserve it for users who wanted that level of control. And it's paired with burst-photo ranking and blur detection - other ways to reduce the mental load of choosing which frame to keep.
What similar-photo grouping actually does in your workflow
You open Culr. You swipe through your camera roll using the familiar keep / delete gesture. When the app encounters a set of visually similar photos, it groups them together. You see them as a cluster - thumbnail strips of the same moment from slightly different angles or exposures.
You can expand the group and review each frame. The app highlights the sharpest frame via sharpness scoring (a bonus Plus feature). Or you can just tap the ones you want to keep and swipe the rest away. No complicated interface. No modal dialogs. You're still in the swipe-and-decide flow; the app just made the candidates visible.
For photographers culling 200 frames from a shoot, this cuts hours. For someone like the customer with 47 birthday-party photos, it turns a confusing wall of near-identical shots into a manageable set of options.
The limit of what grouping can do (and why that matters)
Similar-photo grouping is not burst-photo ranking. It's not the AI Best Shot feature that recommends which frame is technically 'best' based on sharpness, exposure, and composition (that's a Pro feature). It's not a cloud-sync system that backs up your photos somewhere you can't reach them.
What it is: a way to surface candidates so you can make faster, more confident decisions about what to keep. It puts your camera roll in front of you in clusters instead of an endless stream. That clarity matters more than perfection.
We learned this early on. Users didn't want the app to think for them. They wanted it to organize the information so they could think faster. Similar-photo grouping does exactly that.
When you're staring at 47 photos of the same moment, how do you decide which one to keep? That's where grouping lives - not in automation, but in decision-making speed. Have you ever gone back through an old camera roll and wished you'd culled it sooner?