Kepla vs FilterPixel: an honest comparison

The honest answer first: Kepla's picking app is still being built, so you cannot buy it today and this is not a features head to head. FilterPixel is a finished product you can install this afternoon. Here is what it does, what it costs on the pages we loaded, who should pay for it now, and the different bet we are making with ours.

What FilterPixel does today

FilterPixel is a desktop application for Windows and Mac. You point it at a folder of RAW files, it analyzes every frame on your own machine, and it hands back a proposed selection with the obvious problems grouped for you: soft focus, closed eyes, blinks, and the six near identical frames that came out of one burst.

There is also a deeper analysis pass called Deep Cull, and it is metered separately from ordinary projects. Hold on to that distinction when you read the plans, because a single plan line can promise unlimited photos and two Deep Cull shoots at the same time, and those two limits mean very different things.

What happens after the analysis is the part people forget when they shop for any tool in this category. You review what it proposed, overrule it where it was wrong, and send your selection on to Lightroom or Capture One. It does not produce a finished gallery and it does not pretend to. It sorts, then it asks you to sign off. That is true of every AI selection tool on the market, including the one we are building.

We went through the software itself at length in our FilterPixel review, so this page is about a narrower question. How should you weigh FilterPixel against a competitor that has not shipped?

What it costs, from the pages we actually loaded

FilterPixel's pricing moves, and its own pages do not agree with each other. So instead of giving you one number and pretending it is stable, here is exactly what was on the two pages we loaded on August 12, 2026, with links so you can check them yourself.

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

Two pages on one website, two different shapes of offer. Other pages on the same domain have shown other figures again, including a Starter and Pro ladder and a lower monthly headline, which we listed out in the review. Some of that is probably old pages left standing, and some of it is the editing product rather than the culling one.

The practical rule is simple: trust the checkout screen, not a marketing page, and not this article either. Whatever number you land on, weigh it against the hours it is supposed to give back rather than against another subscription, which is what the cost calculator is for.

One thing is not in dispute, and it deserves credit. Four complete projects with unlimited photos and no card is the most generous on ramp in this category. Most trials are clocks that start ticking the moment you install, which is useless if your next wedding is five weeks out. FilterPixel's limit is a count, so it sits there until you have a real shoot to feed it. That alone is a good reason to put it on your shortlist even if you buy something else in the end.

Who should buy FilterPixel this month

If you need software working on the next wedding in your calendar, buy a shipped tool. That is not false modesty, it is the only sensible advice we can give you while ours is unfinished. FilterPixel is a reasonable one to buy, and it is the easiest of the lot to test before you commit.

• You want to try AI selection with no clock and no card. Four full projects on real shoots, at your own pace, is a better test than any review can give you.
• You shoot in bursts rather than steadily. A time limited trial would mostly expire unused. A project count would not.
• You are on Windows. It is a native desktop app on both platforms, and a lot of the category leans Mac.
• Your work is genuinely stuck at the sorting stage. If you are drowning in frames rather than in edits, this is the right category of tool to be shopping in.

Look elsewhere if you shoot volume all year and running cost matters more than trial generosity, because cheaper entry plans exist. We put the field side by side in how to choose culling software, and the two closest rivals get their own page in FilterPixel vs Aftershoot.

The different bet we are making

Full disclosure, in case it is not obvious: we are building a competitor. Read this section knowing that, and read it as a description of a design choice rather than a review of something you can install.

FilterPixel is a desktop program. So are Aftershoot and Narrative Select. That means the job cannot start until you get home, sit down at the desk, plug in a card and wait. For a wedding photographer that is the worst possible moment to be asked to make eight hundred careful decisions, because it is midnight and you have been on your feet since ten in the morning.

Our bet is that the first pass belongs somewhere else. Kepla is being built to make it on your iPhone or iPad, in the passenger seat on the drive home or on the couch the next morning, and to open the same shoot on your Mac when a job needs two nearly identical frames side by side at full size. One shoot, both screens, nothing done twice. We argued that out properly in phone versus computer culling.

Two other differences are worth naming. Kepla marks its picks and nothing else: no file is ever deleted, moved or renamed, every original stays where you put it, and anything it did not pick is one tap away. And it also books the job, which no selection tool does. That part is not a promise, it is finished: the booking page is live and free today while we build the rest.

Whether the phone bet is right is an open question. It might turn out that photographers want the desk. We think the evidence is on our side, but we do not get to score our own exam.

What we refuse to claim until we ship

There are three sentences you will not find on this site, and it is worth saying why.

We will not say we are faster. Speed is measured on real hardware with real RAW files, and until you can run our app on your own laptop and your own phone, any number we published would be a number we chose.

We will not say we are more accurate. Same reason, and a worse one underneath it. Software can objectively check whether eyes are open and whether focus landed. It cannot check whether a frame was the moment. Any single accuracy percentage mixes both and quietly scores only the easy half.

We will not imply the app is available. It is being built, native for iPhone, iPad and Mac, and we will keep saying so on every page until the day it is not true anymore.

This is the right place to talk about FilterPixel's published comparisons, because it is the most enthusiastic publisher of them in the category. FilterPixel says its testing finds 94.7 percent of keepers and 96.1 percent of rejects on a standardized gallery. To its credit, at least one of those pages carries a line disclosing that FilterPixel operates the blog. But those are still its own tests, on galleries it chose, scored against its own definition of a keeper.

That does not make them false. It makes them marketing rather than evidence. And the identical sentence will apply to us. If Kepla ever publishes a benchmark, treat it exactly as sceptically as you treat that one: a starting point, not a finding. Where the honest limits of these models sit is covered in can you trust AI to pick your photos.

One more rule we hold ourselves to, and you should hold every vendor to it. Never take one company's figures for a rival product as fact. Every competitor price on this site comes from that competitor's own page, loaded on the day we wrote it.

How to judge us when the app ships

You should not take our word for any of this, so here is the test we would like to be measured by. It works on FilterPixel today, it will work on us later, and it costs you half an hour.

1. Use a gallery you already delivered. Six or twelve months old, with the delivered set still sitting in a folder. That set is your answer key, and it is the only benchmark scored against your taste instead of a marketing department's.
2. Rebuild the original folder. Every frame from that day, not just the survivors. The tool has to face the same pile you faced.
3. Count exactly two numbers. How many of your real delivered picks did it find, and how many extra frames did it put in front of you that you would never have kept. Those two numbers are the whole review.
4. Then look only at the disagreements. The frames it missed tell you what it cannot see, and they are usually the emotionally loaded, technically imperfect ones. The frames it flagged that you had passed on tell you how long you will still be sitting there at 11pm.
5. Run it twice, in different light. A bright ceremony and a dark reception are different problems, and a tool that only handles one is a tool you stop trusting halfway through a job.

The full method, including how to score it without fooling yourself, is in how to test AI culling accuracy. If you have never timed your own manual pass, do that first, because without a baseline every tool feels fast: culling time benchmarks gives you a number to beat.

When our app ships, run that test on it. If it does not find your work, do not buy it. We would rather lose the sale than have you find out on a job.

So which one, and when

If you need selection software this month, buy FilterPixel or one of its rivals, with a clear conscience. It is a good product from a company that markets harder than the rest of the field, and neither of those facts should decide anything for you. Take the four free projects, run a gallery you already delivered through them tonight, and let your own shoot cast the vote.

If what you are actually looking for is a first pass that happens before you get home, then we are the bet, and the honest position is that you are waiting on us rather than choosing between two shipped products. We make the same argument about the rest of the field in Kepla vs Aftershoot.

In the meantime there is one part of Kepla you can use today, and it is free while we build: the booking page that sends a client your availability, holds the date and takes a deposit. No selection tool in this comparison does that, and it is the only piece of Kepla we will call finished.

FAQ