Adobe's AI for photographers: shipped, announced, and my opinion

Adobe now ships an AI selection pass inside Lightroom, and most of what photographers repeat about it is a year out of date. This is a plain audit: what is on an Adobe page today, what Adobe has only stated as a direction, and where my own guesswork starts. The first pile is linked. The last pile is labeled.

Three piles, and only one of them is fact

Adobe's AI selection feature is called Assisted Culling, Adobe labels it early access, and Adobe's own guide states it is available in both Lightroom and Lightroom Classic. That one sentence retires a line photography blogs repeated for years, including older posts on this site: that Lightroom cannot see a missed focus or a closed eye. It can, and it has for long enough that pretending otherwise is now a credibility problem rather than a rounding error.

The trouble with writing about a roadmap is that vendor talk, shipped software and wishful thinking all arrive in the same paragraph. So this article keeps them in three piles. Shipped means I opened the Adobe page and it is linked in the sentence. Stated means Adobe has said it in its own words, which is not the same as it working the way you hope. Opinion means me, and it is marked. If you want the hands-on version of the feature rather than the map, we cover that separately in our Lightroom Assisted Culling review, and the switches themselves are in the setup walkthrough.

Assisted Culling, feature by feature

In Lightroom Classic you turn it on in Catalog Settings, Metadata, Assisted Culling, or from the drop-down beside the Identity Plate by switching on Auto Analysis. From there the analysis runs in the background and the results arrive as scores you can filter on.

Adobe documents three things it looks for when choosing selects: Subject Focus, Eye Focus and Eyes Open. It also offers rejection criteria for exposure issues, misfires, and documents and receipts, which is a very honest admission about what actually fills a working card. Eye detection has a third state as well as open and closed, a "Can't Tell" bucket for faces where the software will not commit, and you decide what happens to those.

What lands in your catalog is metadata, not surgery. Adobe describes the output as Select and Reject badges you act on in batches: flags, ratings, color labels, or moving the survivors into a collection. If it marks a frame as a reject and you disagree, Adobe's guide says you click the Reject icon and it stops being one. Nothing in Adobe's documentation describes the feature erasing a file from your disk, and it is worth being precise about that, because a lot of nervous forum chatter assumes otherwise.

Cloud Lightroom got a related piece in the June 2026 release: a Faces panel that shows eye sharpness and eyes open scores for each face in a photo. On a family formal with eleven people in it, per-face scoring is a more useful thing to have than a single verdict on the frame, and it is the closest either app comes to how a photographer really checks a group shot.

Auto Stack now groups by how photos look

Stacking is old. What is new is the second criterion. In Adobe's Auto Stack documentation the command sits under Assisted Culling and offers two ways to group: Time, which "groups photos captured within a specific time interval", and Visual Similarity, which "groups photos that look alike based on visual elements". A slider tunes how aggressive the grouping is, and Lightroom promotes what it considers the best photo to the top of each stack. Classic's release notes describe the same feature as stacking based on time and visual similarity.

That matters more than it sounds. Capture-time stacking never understood a burst that paused, and it happily merged the end of the first dance with the start of the speeches. Similarity grouping gets closer to how you think about a shoot: scenes, not seconds.

The risk is that a stack starts to feel like a duplicate list, and it is not one. A run of frames can contain a clean hero shot, a wider frame that gives the room, and a reaction at the edge that carries the story. Collapse it without looking and you have hidden two of the three from yourself.

A stack review routine that takes seconds

• Name the scene before you open the stack, so you know what you are looking for.
• Treat the promoted photo as a reference frame, not a decision.
• Always look at the first and last frame of the run. The gesture usually changes at the edges.
• Keep a variant only when it adds a different emotion, viewpoint or moment, not because it is also sharp.
• Collapse the stack once you have decided, so the day stays readable in the grid.

The analysis moved to import, and that is the big one

Adobe extended the feature so it can run while the card comes in. On Adobe's import page you choose a folder in the Import dialog, select Assisted Culling, and "Lightroom will run an analysis in the background for the selected folder". There is also an option called Don't Import Rejected Images.

Read that option carefully before you use it on a job that matters. The files are not deleted, they stay on the card or the drive exactly as they were, but they never enter the catalog, which means they never enter your review either. You are trusting the model's judgment before you have looked at a single frame. On a corporate headshot day that is a reasonable trade. On a wedding, I would leave it off until the tool has earned it on jobs you already delivered.

Two more shipped items belong on the list, both from Classic's what's new page. Version 15.4 in June 2026 added duplicate image auto-detection plus AI metadata filters and Smart Collections, so machine scores can drive a saved view rather than a manual pass. Version 15.5 in August 2026 was an editing release: crop opacity, feather and edge sliders for masking, render to DNG. No new selection AI in it, which is itself a useful data point about pace.

What is real, what Adobe merely says, and what is me talking

Adobe pages checked August 2026. Anything not in this table did not make it into this article.

The pattern is clear enough. Everything shipped is a measurable property of a single frame, or a similarity relationship between frames. Nothing shipped is about meaning, and Adobe is not claiming otherwise.

"Optimized for portraits and headshots" is a load-bearing sentence

Adobe states plainly that the current feature set is optimized for portraits and headshots. Take that seriously, because it maps almost perfectly onto where the feature will delight you and where it will annoy you.

It fits work with a narrow technical standard: school portraits, corporate headshot days, family formals, product runs, a burst of a couple walking toward you in even light. In all of those, sharp eyes and open eyes really are close to the whole brief, and a tireless machine applying that test to frame 1,800 as carefully as frame one is a genuine gain.

It fits badly where the frame is ambiguous on purpose. A dark reception, a dragged shutter, a silhouette, a face in profile, glasses catching a flash, a hand across the eyes, a dance floor at 1/15s. Those are not failures, they are decisions, and the scores that make the feature useful in a headshot studio have no way to tell the difference. The categories that trip up every tool in this market, not just Adobe's, are set out in what AI culling misses, and the feature's specific edges are in Assisted Culling's limits.

The practical version is a rule about thresholds, not a rule about trust. Let the filter run hard on the predictable parts of a job. Loosen it, or skip it, on the twenty minutes of the day that people will still be looking at in thirty years.

Where I think this goes next, clearly labeled as guesswork

Adobe has not published a roadmap for these features beyond the release notes above, so nothing in this section is a promise from anyone. It is my read, and you should discount it accordingly.

I think the catalog becomes decision-aware rather than storage-aware. Smart Collections that already accept AI metadata filters are a small hint of a larger direction: your library stops being a filing cabinet and starts being an assistant with opinions. For photographers who live entirely inside Adobe, that is a convincing reason never to open a second app, and it is the strongest competitive move Adobe has made in this category.

If that happens, the discipline problem arrives with it. When machine labels and your labels share one vocabulary, you can no longer tell your judgment from the software's. The fix is boring and works: decide what your own marks mean and never let an automated score write into them. Flags mean reviewed by a human. Stars mean editing candidate. A collection means confirmed for the client. We compare the options in ratings, flags and labels.

Teams feel this first. An associate reads a stack as "pick one", a lead reads it as "compare these four", and an editor reads a star as "deliver this". Before next season, write down three answers: which folder holds the untouched originals, which mark means a confirmed selection, and who is allowed to pull back a frame the software pushed down. Those answers outlive any feature release, and they are the same ones that make a second shooter handoff work.

A sane way to try this in the next two weeks

You do not need to redesign your business around a feature in early access. Do it in this order. Standardize folder names and backups first, because every workflow improvement downstream depends on them. Then run Assisted Culling over a job you already delivered, where you know the right answer, and compare its picks against your gallery. Watch the disagreements, not the agreements. That is the whole test, and we wrote the half hour version of it in how to test AI culling accuracy on your own shoot. Then change one habit at a time.

Cost barely enters into it. If you already pay Adobe, this is included: the photography plan with 1 TB is $19.99 a month and Lightroom on its own is $11.99, and Assisted Culling is not an extra line on the bill. The honest question is not price, it is where the work happens. Everything above still requires you home, at the desk, with the card in the reader, which is the part of the evening most photographers actually resent. We put numbers on that tradeoff in phone versus computer culling.

FAQ