Culling group photos: getting every eye open
Twenty two people. Nine frames. In every single one, somebody has their eyes shut, and it is never the same somebody twice. This is the least glamorous problem in the whole gallery and it eats more time per delivered photo than anything else you shoot. Most of the fix happens before you ever open the folder.
Why the frame with everyone open often does not exist
The reason large formals fail is not that your subjects are uncooperative. It is arithmetic. One person has a small chance of blinking during any given exposure. Two people have that chance twice over. By the time you have twenty faces the chance that at least one pair of eyes is shut has become the likely outcome, and three frames of a twenty person group is not a plan, it is a wish.
Two CSIRO researchers, Nic Svenson and Piers Barnes, worked this out and won the 2006 Ig Nobel Prize in Mathematics for it. Their rule, as reported at the time, is for groups of fewer than twenty people: divide the number of people by three in good light or with a good flash, and by two when the light is poor. That is how many frames to take.
Barnes was cheerfully honest about it, saying they were proud to have made a gross simplification of complex physiological and psychological factors backed up with no empirical data. Take him at his word: it is not a guarantee. What it gets undeniably right is that the number of frames a group needs scales with the number of people in it. Almost nobody shoots that way. Most photographers fire the same three or four frames whether it is a couple or a rugby team.
| People in the group | Good light or good flash | Poor light |
|---|---|---|
| 6 | 2 frames | 3 frames |
| 9 | 3 frames | 5 frames |
| 12 | 4 frames | 6 frames |
| 15 | 5 frames | 8 frames |
| 18 | 6 frames | 9 frames |
The rule is Svenson and Barnes, CSIRO, 2006, applied here and rounded up. The researchers offered it only for groups under twenty. Above that, shoot more and expect to work harder at the picking.
The picking is decided at the shoot, not at the desk
You cannot pick a good frame out of a set that does not contain one. Four habits, and none of them cost you time on the day.
- Count the group and shoot to the count. Six people is a couple of frames. Eighteen in a dim reception room is closer to nine. It feels excessive in the moment. It is the difference between a two minute pick and a twenty minute one.
- Keep the framing identical across the run. This is the big one. If you recompose between frames, the frames stop being interchangeable, and interchangeable is what makes both fast comparison and any later combining possible. Lock it, shoot the run, then recompose deliberately and shoot a second run.
- Shoot the run as a run. Consecutive frames, a beat apart, not scattered across two minutes of chatting. Grouping tools and your own eye both work on continuity, and a set spread over ninety seconds arrives at the desk as five orphans.
- Call it, then shoot again after the laugh. Everyone knows the countdown trick. The part people skip is a second short run five seconds later, once the group has relaxed out of the braced pose. That run is frequently where the actual pick comes from.
Watch the back row specifically. Furthest from your light, most likely to be squinting, and smallest in the frame, which is also where software stops helping.
Why an eyes-open score means less as the group grows
AI culling handles a single portrait well. It finds the face, finds the eyes, produces a number. On a formal with sixteen faces it has to produce one number for the whole frame, and there is no honest way to do that.
If the tool takes the worst face, one cousin in the back row sinks the frame where the couple look wonderful. If it averages, one closed pair gets diluted by fifteen open ones and you find it during editing. If it counts closed eyes against a threshold, someone has decided how many blinks are acceptable before you have seen the photo. Most tools do not tell you which they do. We took that apart properly in how blur and blink detection actually work.
Then there is resolution. A back row face in a wide formal can be a hundred pixels tall. Detectors need more than that to tell a blink from a downward glance. Add sunglasses, glare on lenses, a head turned three quarters, a face half behind a shoulder, and the score is being computed on partial information without saying so.
None of this is a scandal. Adobe explicitly describes its Assisted Culling as optimized for portraits and headshots. Use the software for what it is genuinely good at here: gathering the run into one place and offering a starting candidate. Then check the faces yourself, because on a formal, that check is the job.
A reading order for a frame with twenty faces
The failure mode is scanning. Your eye lands somewhere, wanders, catches a closed pair, jumps to the next frame, and you have no idea whether you checked the man on the far left in either of them. Then you start over. Replace scanning with a fixed route.
| Pass | What you look at | What ends the frame |
|---|---|---|
| One | Eyes only, back row left to right, then each row forward | A closed pair on anyone who matters |
| Two | Expression, on the two or three people the photo is for | A grimace, a mid word mouth, a braced smile |
| Three | Bodies: hands, a turned head, someone leaning out of the group | A person who does not read as themselves |
| Four | Edges, gaps, the background, the tilt | Almost nothing, this is crop territory |
Three things make the route work. Use the same path on every frame, so a face you skipped is skipped consistently and you notice. Record what you find as a count and a position, not a name: "two closed, back left and front right" beats trying to remember faces. And hold a champion rather than ranking the set, comparing each new frame only against the current best and never reopening a settled comparison. That knockout method, and the reason it is so much faster than ranking, is the subject of our guide to choosing the best frame from a burst.
Pass one usually leaves you with two or three survivors out of nine. Everything after that is quick, because you are no longer looking at twenty faces, you are looking at three.
Head swaps: name the cost, and the disclosure question
Sooner or later you get the frame where everything is right except one uncle, and the frame beside it where only that uncle is right. Plenty of photographers combine the two, taking one face from the second frame into the first. On posed group work it is ordinary, long standing practice.
This article is not going to teach you the retouch. How you blend two frames is a Photoshop question with better teachers elsewhere. What belongs here are the three things that go wrong at the picking stage.
A picked frame that needs a swap is two files, not one
The hero is worthless without its donor, and the donor is by definition a frame that scored worse, more closed eyes, a duller expression, which is exactly why you were not going to pick it. So flag both, as a pair, in the same pass. Two things break this. One is a workflow that files low scoring frames off into a rejects folder, because your spare part is now somewhere you have stopped looking. The other is your own memory in three weeks, when the hero reaches the edit queue and the donor is one frame among 3,000.
It is an editing bill you sign while culling
Every swap is minutes of work you have just bought for a future evening, so price it like any other extra edit. Cap yourself: if a setup needs three swaps, pick a different frame or accept that the run failed. Decide at the pick, not at 1 AM.
Some clients and some genres treat this as a disclosure matter
In wedding and family portraiture, a swapped face on a posed group is craft and almost nobody thinks twice. Elsewhere it is a different thing entirely. News, documentary and most editorial work draw a hard line between a photograph and a composite. Photo competitions usually have explicit rules. Some corporate, institutional and legal work does too, and so do some individual clients who simply want to know. Know which room you are standing in. If you are not sure, ask before the shoot rather than after a complaint, and if it could ever matter, say so in your contract. It is one sentence, and it is a much easier sentence to write in advance.
Decide who has veto power before you open the folder
In a twenty two person formal, somebody is always slightly wrong. If everyone gets a veto, no frame survives and you will circle the run until you give up and pick at random.
So rank the group before you look. The people the photo is for sit at the top: the couple, the honoree, the grandmother who may not be at the next wedding. Everyone else is below the line, and below the line the standard is acceptable, not perfect. A back row cousin with lazy eyes is not a reason to lose the frame where the grandmother looks radiant, and no client has ever noticed the cousin.
The same idea takes a different shape in small family sessions, where you rank by who is controllable rather than by who the photo is for, because the toddler will give you what the toddler gives you. That version is worked through in our guide to culling family sessions. And in newborn work the problem inverts again: nearly every frame is technically fine and the separation has to come from somewhere else entirely, which is the subject of culling a newborn session.
Write the ranking down if it helps. Two names is usually enough.
One frame per configuration, and the honest exception
Formals are a checklist deliverable. The family wants each combination once: both families, her side, his side, the grandparents, the siblings. They do not want four versions of her side with different cousins blinking, and delivering them reads as padding rather than generosity.
The choice research backs this up. The classic Iyengar and Lepper jam study found that more options lowered both purchases and satisfaction, and a formals section is the easiest place in a wedding gallery to hand someone thirty near identical images at exactly the moment they are trying to choose a print. Sizing the gallery as a whole is covered in how many wedding photos to deliver.
The exception is narrow. Keep a second frame when it is your spare part for a swap, or when the group genuinely gave you two moments, the formal one and the one where they all cracked up. Keeping it does not mean delivering it, and the alternates stay in your files either way, which is the point of what to do with the photos you do not pick.
How we are approaching this at Kepla
Kepla checks every frame against clear criteria, sharp focus, open eyes, expression, near identical neighbors, and hands you a shortlist. From a 2,000 photo wedding that is roughly 800 picks, in minutes. On formals, treat its proposal as a starting candidate rather than a verdict.
The part that matters most on group work is what it does not do. Kepla never deletes, moves or renames a single file. It only marks what it picked. Every frame in the run stays exactly where it is, one tap away, so the donor frame you need for a swap is still sitting next to the hero six weeks later instead of in a rejects folder you emptied.
The first pass happens on your iPhone or iPad on the drive home. Then, when a formal needs twenty faces checked at full size, the same shoot opens on the Mac and you carry on: side by side, zoomed in, every frame still there. One shoot, both screens, nothing redone.
Stated plainly: the picking app is still being built for iPhone, iPad and Mac. The booking page is live and free today. Founding members get the picking app the day it ships.
FAQ
How many photos should I take of a group so nobody is blinking?
CSIRO researchers Nic Svenson and Piers Barnes proposed dividing the number of people by three in good light, or by two in poor light, for groups under twenty. Six people means two frames, eighteen in a dim room means nine. They admitted it was a rough simplification, so do not treat it as a guarantee. The useful part is that frame count should scale with group size, and most photographers shoot a flat three regardless.
How do I cull a group photo where someone always has their eyes closed?
Rank the group before you look. The people the photo is for, the couple, the honoree, the grandmother, must be right. Everyone else only has to be acceptable. Then read every frame along the same fixed path, back row first, counting closed eyes by position rather than by name. Two or three survivors usually emerge on that first pass, and you judge expression only among those.
Is it okay to swap a head from another frame into a group photo?
In wedding and family portraiture it is ordinary practice for posed groups. In news, documentary, most editorial work and photo competitions it is not, because those treat a combined image as a composite rather than a photograph, and some corporate clients and individuals feel the same way. Know which context you are in, ask before the shoot if you are unsure, and cover it in your contract if it could matter.
If I swap a face between two frames, which files do I keep?
Both, flagged as a pair, decided in the same pass as the pick. The hero frame is useless without the frame you are borrowing from, and that donor is by definition one that scored worse, so any workflow that files low scoring frames somewhere separate has quietly taken your spare part away. Weeks later, hunting for one specific frame among thousands is a job you do not want.
Does AI culling work on large group photos?
It works well on one face and gets unreliable as faces multiply. The software must collapse sixteen eye scores into a single number for the frame, and every way of doing that is wrong somewhere. Back row faces are often too small in the frame for reliable eye detection anyway. Adobe describes its own assisted culling as optimized for portraits and headshots. Let it group the run, then check the faces yourself.
How many formal group shots should I deliver?
One per configuration in almost every case: both families, each side, the grandparents, the siblings. Clients want each combination once, not four versions with different people blinking. Keep a second frame when it is your spare part for a swap, or when the group truly gave you two different moments. Keeping a frame and delivering it are separate decisions, and only one of them is the client's problem.
Check twenty faces without checking them twice.
Kepla for Mac clears the obvious misses from a card, names the reason on every frame it sets aside, and leaves the choosing to you. Nothing is ever deleted, moved or renamed. Free through the private preview · the first hundred photographers keep it at $99 a year.