Culling trends, 2026 into 2027: what actually changed

Most year in review pieces are guesses with a date on them. Two rules here. Anything written in the present tense links to a page I opened in August 2026. Anything about next year is labeled opinion and you can skip it. And where the honest answer is that nothing moved, that is what it says.

This is about the decision, not the software

We already published the market side of this year: which tool shipped what, which prices moved, and where they all sit now. That is in photography workflow trends 2026, including the pricing comparison, and there is no point repeating it here.

This piece is about something the tool coverage tends to skip. The interesting change in 2026 was not which app you bought. It was that selection stopped being the shapeless gap between shooting and editing and started being treated as a stage of the job, with an input, a decision, a handoff and a defined end. That shift happened in studios that changed software this year and in studios that changed nothing at all.

Three things are worth saying up front about how this was put together. Vendor accuracy comparisons are not used as evidence anywhere below, and neither are any vendor's published figures for a rival product. No number appears unless it came from a page linked in the text. And the last section is opinion, marked as opinion, because the honest label for a claim about 2027 is a guess.

The volume problem finally got named

Modern cameras are extremely good at generating insurance. Large sensors, deep buffers, silent shutters and subject tracking make it easy to come home from a wedding, a conference or a ninety minute portrait session with thousands of files. That capability is genuinely valuable when the action is unpredictable. It also quietly creates the assumption that every frame deserves equal attention later.

A full wedding day still commonly produces 2,000 to 4,000 RAW frames captured against 400 to 800 delivered. Those are typical industry ranges rather than a measured study, and they have not moved. What changed in 2026 is how many photographers started treating the capture count as a variable they control rather than a fact they receive.

The photographers who got time back did not shoot less across the board. They got specific about when a burst was solving a real risk. A first look, a kiss, a dance move, a toddler mid laugh or a fast sports sequence can justify a run of frames. A settled portrait of a person who is not moving generally does not. Capture discipline remains the cheapest selection tool available, and it is the only one that also reduces your backup time, your storage bill and the number of thumbnails a human brain has to judge.

If you want to know whether this applies to you, count before you change anything. There is a method for that in the culling time audit.

AI settled into the first pass, and stayed there

The useful framing that emerged this year is narrow: these tools produce a candidate collection, not a delivery folder. They are good at the mechanical scan and they are not making the call.

What they actually do is now easy to check. Adobe's help page for Assisted Culling in Lightroom Classic describes filters for subject focus, eye focus and eyes open, plus rejection filters for exposure issues, misfires, and documents and receipts. The same page says you can cull at the time of import, and it describes the result as flagging, rating, labeling and collecting rather than deleting anything. It also still carried the Early Access label when I read it on August 12, 2026, so treat the status as in transition. There is a setup walkthrough in our Assisted Culling setup guide.

Cloud tools describe a similar shape from a different place. Imagen's culling page says the service is cloud based so it does not use your computer's processor, and lists face recognition for grouping subjects and duplicates, blur detection, closed eye detection and a kiss recognition exception for the blinks that are actually keepers. Aftershoot's pricing page describes offline access and no limit on how many images you can cull, which is the local end of the same idea.

Read those descriptions closely and the boundary is right there in the vendors' own words. Every capability listed is technical: focus, exposure, eyes, duplicates, misfires. None of them is about whether the moment was worth having. A model can rank the sharpest face in a family group and miss the frame where the family is actually connected, or reject the street photograph whose timing is the entire point. It also cannot know a client's shot list, a privacy restriction or the thing the couple told you in the car. We keep a running list of that gap in what AI culling misses.

The selection brief moved before the shutter

The most practical operational lesson of the year is that culling starts before the shoot, not after it. The studios with fast, defensible edits are usually the ones that wrote down what the job needs before anyone picked up a camera. "Send me the good ones" is not a brief. "One retouched headshot per employee, square and vertical crops, a second option for the executive team" is.

A workable brief answers five questions, and it fits on half a page.

1. What does the client receive, and how many finished images are expected?
2. Which moments, people, products or views are non negotiable?
3. Where do the images end up: print, web, social, a layout, an archive, a license?
4. What level of retouching and what turnaround are included?
5. Who is allowed to approve an exception or an extra selection?

This prevents the classic late discovery that your favorite frame cannot take the crop the layout needs, or that a required group was never photographed in a usable way. It also gives an assistant, a second shooter or a piece of software the context it cannot invent for itself. That is the same discipline as writing your culling criteria down, pointed at one job instead of your whole business.

Selection and retouching came apart

One of the most expensive habits in photography is editing before selection is settled. Every frame that gets a preset, a crop and a local adjustment before anyone decided it was going out is time spent polishing files that will live in an archive forever.

Through 2026 more workflows split into visible stages: ingest and backup, technical rejection, candidate selection, approval, then retouching and export. The split does not have to be rigid. A quick base correction is often the only way to judge a difficult backlit frame. The working rule is to do just enough processing to make the decision, then hold the skin work, the product cleanup and the color finishing for confirmed selects.

The reason this counts as a trend rather than advice is the handoff. Once a studio has more than one person touching a job, selection status has to live somewhere other than in the photographer's head. Flags, collections and smart collections do it in Lightroom, sessions and albums do it in Capture One, and a fast browser does it for the first technical pass. Whichever you use, the test is the same: can someone else tell what has been decided without asking you?

One score stopped meaning one thing

The year also killed off the idea that a keeper is a keeper. A corporate headshot needs reliable expression, accurate identity and consistent crop. A drone frame needs coverage of the brief, sharpness under movement and light that means something. A boudoir set needs privacy aware selection above all of it. A street photograph can survive on timing that would fail a product catalog outright.

This is obvious when written down, and most generic workflows still ask you to rate everything against the same five star ladder. The fix is a short genre specific check rather than a universal one.

The rating is only the final signal. The thinking that produces it should change with the work, and that is the part no setting carries over between genres.

Three things that did not move at all

The workflow level version of this list, covering where the work happens and what the tools still cannot do, is in the workflow trends piece. These three are specific to selection.

The number of decisions is the same. A first pass that proposes 800 candidates out of 3,000 frames has not reduced how many images you look at. It has changed your job from deciding to reviewing, which is faster per frame and is not free. Reviewing several thousand of somebody else's judgments has its own kind of fatigue, and anyone promising otherwise is selling something.

The criteria are still yours to supply. Read the vendor descriptions above again. They detect focus, exposure, eyes and duplicates. Not one of them knows what this client asked for. That gap is the same size it was a year ago.

Nobody has measured any of it. The most quoted time figure in this industry comes from a UK survey of more than 300 wedding photographers by Your Perfect Wedding Photographer, reported by PetaPixel, which put shooting at 4 percent of working time, editing at 55 percent and culling at 11 percent. That was published in 2020, before any of this software was mainstream, and the article itself warns about the sample size and the fact that it is UK only. No comparable follow up appears to exist. So every claim about how many hours AI has given the profession back is resting on a six year old baseline with nothing after it. We keep that argument current in where a photographer's time actually goes.

What I think 2027 does, labeled as opinion

Everything above is checkable against a link. Everything here is my read, and you should discount it accordingly.

Opinion: the brief becomes the product. If the technical pass is close to solved and available inside the editor most people already pay for, the remaining value is in describing what this job needs and having the software respect it. I think that is where the interesting work goes next, and it is a harder problem than blink detection because the input is a sentence rather than a pixel.

Opinion: the constraint that bites is location, not accuracy. Detection is good enough that the annoying part of a working week is not the model's judgment. It is that the whole process is chained to a desk on a Sunday. Moving the first pass somewhere else looks more valuable to me than another two points on blink detection. That is the bet Kepla is making, and the status is exactly this: the picking app for iPhone, iPad and Mac is still being built, and it will pick and shortlist without ever deleting, moving or renaming a file. The booking page is live and free today. I am not going to predict dates beyond that.

Opinion: the ceiling is closer than the marketing suggests. A model can tell you which frame is sharp. It cannot tell you which frame the bride's father will cry at. I do not expect that gap to close in 2027, and I would be suspicious of anyone who says it will.

A more useful thing to do than reading predictions: audit one job you already finished. Count the frames captured, the minutes spent on the first pass, the minutes spent on the fine selection, the images delivered, and the exact point where you stalled. Then change one thing, on one type of session, and keep it only if it holds. Slower than a software switch, and it is the version that survives a busy season. If you want the wider picture of what the year did to the business rather than the workflow, that is in the state of the photography business in 2026.

FAQ