Culling: how to get from 1,200 frames to 40 without losing a day

Decide the number before you open the folder

Write down how many images you are delivering before you look at a single frame. Sixty for this portrait session. Four hundred for this wedding. Twelve for this editorial. If your contract states a number, that is your number. If it does not, pick one anyway.

Without a target, culling becomes an open-ended aesthetic exercise and you will deliver whatever you happen to still like at 1am. With a target, every decision becomes comparative: is this frame better than the weakest frame currently in the set? That question has an answer. Do I like this frame? does not.

A target also protects the client. Handing over 900 images is not generosity, it is transferring your unfinished work to someone who is less equipped to do it. Clients who receive 60 strong images look at all 60. Clients who receive 900 look at 30 and feel vaguely tired.

Pass one: reject on technical failure only

The first pass is fast and mechanical. You are not choosing favourites. You are removing frames that cannot be delivered no matter how good the moment is: missed focus on the subject's near eye, motion blur you did not intend, eyes fully closed, someone walking through the frame, a fired flash that did not, obvious duplicates within a burst.

Move quickly and do not agonise. If you find yourself thinking about a frame for more than about two seconds in pass one, leave it in — that hesitation means it is a judgement call, and judgement calls belong in pass two.

Do not zoom to 100 percent in this pass except to check critical focus, and even then only on frames you actually expect to use. Pixel-peeping every frame is the single biggest time sink in culling and it tells you almost nothing about whether an image is worth delivering.

Pass two: rank on moment, expression and the brief

The second pass is the real work and it is judgement, not technique. Within each cluster of similar frames, you are picking the one where the expression lands, where the hands are doing something coherent, where the background is cleanest, where the moment reads.

Hold the brief in your head while you do it. A wedding gallery needs coverage — the people who were there, the details that were paid for, the arc of the day — not forty of your best frames of the same two photogenic guests. A commercial job needs the shot the art director described, even if the frame you love more is the one three clicks to the left.

Work in comparison mode where your software offers it. Frames are almost impossible to rank in isolation and easy to rank in pairs. Two up, keep the winner, next challenger, repeat. This is faster and produces better sets than star-rating each frame on its own.

Do not edit while you cull

The most expensive mistake in post-production is developing a frame you have not yet decided to deliver. It feels productive. It is not — it commits five minutes to an image with a fifty percent chance of being cut, and worse, it makes you reluctant to cut it afterwards. Nobody deletes a frame they just spent time on.

This applies to AI tools with particular force. It is tempting to run a promising-but-cluttered frame through an editor mid-cull to see what it could be. Resist it. Foxi AI and similar tools work one image at a time and cost credits per edit, so a mid-cull experiment on a frame you then reject is money and attention spent on nothing. Finish the cull first, then decide which of the surviving frames need more than a develop.

The corollary: a frame should never survive the cull because you think you can fix it. It survives because the moment is right and the technical faults are within normal correction. Fixability is a pass-three question.

Why no AI tool can do this stage for you

Automated culling tools can sort on sharpness, on eyes-open, on faces detected, on duplicates. That is genuinely useful for pass one, and if a tool speeds up your technical rejections, use it.

Pass two is not automatable in any meaningful sense, because the criteria are not in the pixels. The tool does not know which guest is the client's dying grandfather, that the brief specified the product in use rather than on a plinth, or that the client explicitly asked for no images from the left side of the room. An AI photo editor is a tool for changing an image you have already chosen. It has no access to why you chose it.

Treat that limit as freeing rather than disappointing. Culling is the part of the job where your judgement is the product. It is the reason the client hired you rather than renting a camera.

A repeatable session structure

Culling goes badly when it is open-ended, so give it a shape and a clock. The structure below is a reasonable starting point that you can adapt to your own volumes; the important part is that each block has an end.

Set a timer per block and stop when it goes. An imperfect set delivered on Tuesday beats a perfect one delivered next month, and in practice the difference between your 40 best and your 40 second-best is invisible to everyone but you.

• Ingest and verify copies, then walk away from the machine for ten minutes
• Pass one: technical rejects only, no zooming, no ratings beyond reject
• Break — the fatigue point is real and shows up as inconsistent standards
• Pass two: compare in pairs within clusters, promote winners toward your target
• Trim to the number: cut the weakest survivors rather than defending them
• Only now decide which frames need develop work, retouch or a salvage attempt

Frequently asked