AI culling explained simply: what it actually does

You do not need to understand how AI culling works to decide whether to use it. You need to know two things: what it will do to your photos, and what it will do to your evening. This is the plain version. No jargon, no technical detour, and no pretending it is magic.

What it does, in four steps

You copy the photos off your card into a folder. You point the software at that folder. It looks at every frame and hands you back a shorter pile, the ones it thinks are keepers. You go through that shorter pile and say yes or no.

That is the whole thing. It does not edit your photos. It does not send anything to your client. It does not reorganize your hard drive. It does one job: it reads a big pile and proposes a smaller one, and you get the final say on every frame in it.

The reason anyone bothers is arithmetic rather than novelty. A wedding day is typically 2,000 to 4,000 frames, and somewhere between 400 and 800 of them get delivered. Somebody has to look at the difference. For years that somebody was you, at night, with the laptop hot on your knees. If the word culling itself is new, start with what photo culling actually means.

What it does to your photos: nothing

This is the question that stops most people, so here it is plainly. The software reads your photos the way you read a book. It does not write on them. The picture inside each file is not changed, not squashed, not re-saved. Your folder still has every frame you shot, under the name you gave it.

What it produces is an opinion sitting alongside your photos. Usually that is a star rating or a colored label, the same kind you would apply yourself, or just a mark inside the app saying this one made the shortlist. That is all a pick is.

One thing worth checking before your first paying job. A few tools offer a setting that tidies the frames they did not pick into a separate folder. It is optional, and you want to know whether yours has it before you run a wedding through it rather than after. Aftershoot, for example, says in its own culling FAQ that it is non-destructive, that it never deletes images, and that every photo stays reachable in a rejected view. Look for that sentence for whichever tool you pick.

And a frame that misses the shortlist is not gone in any sense that matters. It scored lower on a few measurements, which is not the same as being a bad photograph, and it is sitting right where you left it. There is a whole second gallery in that pile, which we wrote about in what to do with the photos you don't pick.

What changes about Sunday night

The honest answer is that it does not skip a step, it changes what the step feels like.

Without it, you open a folder of three thousand frames and make three thousand decisions, most of them about whether something is in focus. Your standard at frame 200 is not your standard at frame 2,800, because saying no takes more energy than saying maybe, and eventually everything becomes a maybe.

With it, you open a shortlist of a few hundred and do a different job: checking someone else's proposals and overruling the ones you disagree with. That is faster, but it is not nothing, and anyone selling it as one click is selling you something.

It is worth knowing how much of your working life this step takes. A UK survey of more than 300 wedding photographers, reported by PetaPixel, put culling at 11 percent of working time against 4 percent actually spent shooting. That survey is from 2020, before these tools were any good, and the shape of it still holds: the picking sits between the wedding and everything else, and until it is done nothing else can start.

The part it cannot do

The software is good at the boring half. It can tell you which frame in a burst is sharpest, which faces have their eyes shut, and which six photos are versions of the same second. It does that at the same standard on frame 3,000 as on frame one, which is the one thing you genuinely cannot do at midnight.

It has no idea what any of it means. It does not know which woman is the bride's mother, that the man crying at the back is her father, or that the couple asked you in March to get a photo with the dog. It cannot tell a deliberately soft, gorgeous, slow shutter first dance from a mistake, and it will often flag the frame you are proudest of. That is not a bug you can train away, it is the edge of what measurement can do.

So the split that works is simple. Let it do the mechanical pass. Keep the editorial one. If you are curious about which measurements it makes, we went through them in how AI picks your best photos, and the machinery around that is in how AI culling software works. If you are not curious, you have lost nothing by skipping both.

It will get things wrong, and that is useful

Do not put a client's wedding through a new tool cold. Take a shoot you already culled by hand months ago, run it, and compare.

Then ignore everything it got right and look only at the disagreements. There are two kinds. Frames it picked that you would not, which is mildly annoying and costs you a few seconds each. And frames you picked that it passed over, which is the important number, because those are the ones that would have quietly gone missing from your gallery if you had trusted it blindly.

After two or three shoots you will know its personality: most tools are strict about focus, nervous about closed eyes, and hopeless with intentional blur. Once you know that, you know where to look, and the checking gets much faster. There is a proper method for this in how to test AI culling accuracy, and the wider trust question in can you trust AI culling.

One habit worth keeping from day one: never let the software be the only place a decision is recorded before you have exported. Your backups matter more than your picks, always.

What it costs

Less than most photographers expect, and every one of these has a free trial that does not need a card, so you can answer the question with your own photos rather than a review.

Prices checked August 2026 from each vendor's own pricing page.

Two things to notice. Aftershoot and Narrative charge the same whatever your volume, so a busy August costs no more than a quiet February. And Photo Mechanic is in a different category on purpose: it makes you faster without deciding anything, which some photographers prefer and which is a perfectly respectable answer. If you want to see what your whole software bill adds up to, we built a cost calculator for that.

Who this actually helps

It helps most if you shoot volume and the backlog is the thing keeping you up. Weddings, events, sports, big family sessions, anything where you come home with thousands of frames and dread the folder. That is where the mechanical half is enormous and the software takes almost all of it.

It helps least if your shoots are small. Sixty frames from a portrait session is not a sorting problem, it is a ten minute job, and no tool improves on that. It also helps less if your style leans soft, dark or deliberately blurred, because those are exactly the frames a strict tool argues with, and you will spend your time overruling it.

And there is a group who simply enjoy the picking, who find the second pass through a wedding is where they see the story. If that is you, nothing here is an argument against it. Use the tool to group the near identical frames and leave the choosing alone. There is a longer version of this decision in is AI culling right for you.

What Kepla is doing, honestly

We are building one of these, so treat this as interested rather than neutral.

Our bet is about where the work happens. Every tool above runs on a computer, so the earliest you can start is when you get home and plug the drive in, which on a Sunday night is nobody's best hour. Kepla does the first pass on your iPhone or iPad, on the sofa, and then the same shoot opens on the Mac when you want a big screen for the close calls. Nothing gets redone in between.

And it only ever marks its picks. Nothing is deleted, moved or renamed, every original stays exactly where you put it, and any frame it passed over is one tap from joining the shortlist. If disagreeing with your software is hard work, you stop disagreeing, and then it is making your choices for you.

Plainly stated: the picking app for iPhone, iPad and Mac is still being built and has not shipped. What works today is the Kepla booking page, which holds a date and takes a deposit, and it is free while we build.

FAQ