For most wedding and event photographers, start with an AI culling tool such as Aftershoot, Narrative Select, FilterPixel or Lightroom Classic's Assisted Culling, then add a style-trained editor, and leave retouching until last. Culling needs no training and its misses are easy to catch. Confirm that order by timing one paid gallery stage by stage.
Most roundups rank tools. A working photographer needs a sequence, because the three stages fail in different ways: a culling error loses a moment, an editing error breaks consistency across the gallery, and a retouching error changes a face the client knows. A culling miss is cheap to catch, editing drift shows up when you scroll the gallery, and a retouching slip may only appear when a relative enlarges a print, which is why retouching goes last.
Start with a timesheet from one paid gallery
Pull up the last wedding you delivered and write down the hours by stage, from your records or an honest estimate. Here is an illustration for a single-photographer wedding: 4,200 frames shot, 650 delivered, six weeks promised.
| Stage | Hours | Share | What goes wrong if the AI errs | Who notices |
|---|---|---|---|---|
| Import and backup | 1.0 | 4.5% | Nothing AI-specific | Nobody |
| Cull | 7.0 | 31.8% | A missed moment, such as the ring exchange | The couple, when it is absent |
| Edit | 9.5 | 43.2% | The look drifts between lighting scenes | The couple, scrolling the gallery |
| Retouch | 3.0 | 13.6% | Plastic skin or a changed face | The couple and their families |
| Export and gallery | 1.5 | 6.8% | Nothing AI-specific | Nobody |
| Total | 22.0 | 100% |
Editing is the biggest block, so it looks like the obvious first target. It is not, for two reasons. An editing tool has to learn your look from your past edits, so it asks for setup before it saves a minute, and its mistakes are scattered across hundreds of frames. Culling needs no training, and you can review its picks and rejects in a couple of hours. On this timesheet the order is cull first, edit second, retouch last, with AI touching only the hero portraits at the end.
A different business gets a different order. A headshot studio that finishes 40 faces a day should start with retouching. A same-night event photographer should start with culling plus a generic batch edit. Software picks by genre and volume are covered in the tutorial on AI software for photographers by genre; this one is about the sequence inside a paid job.
Culling: test it where a miss would hurt
Four culling tools are worth a test. Aftershoot's Select sorts thousands of photos into selects with assisted or fully automatic modes. Narrative Select sorts, assesses and ranks, then leaves the final call to you, with an AI First Pass and a Close-Ups panel for checking faces. FilterPixel has a Basic Cull and a genre-aware DeepCull that gives each photo a score and a reason, and says it never removes an image. Lightroom Classic has Assisted Culling built in, which selects for subject focus, eye focus and open eyes and rejects exposure problems, misfires and photos of documents or receipts. Its criteria are about faces and eyes, so it suits weddings and portraits better than landscapes or products.
Whichever you try, run it in a mode that only rates and rejects, never one that deletes. Rejected frames should stay in your catalogue until you choose to delete them, and that is the behaviour to insist on from any tool.
Now the test, using the same illustrative wedding. You kept 650 frames from 4,200. The tool picks 720, and 585 of them are frames you kept.
- Recall: 585 of your 650 keepers, so 90%.
- Precision: 585 of its 720 picks, so about 81%.
- The misses: 65 frames you kept were not picked. Sixty-two are near-duplicates of frames it did pick. Three are shot-list moments: the second angle of the ring exchange, the grandmother's portrait and a laugh in the middle of a speech.
Overall recall is respectable, but three missed moments make this a fail for unattended use and a pass for assisted use, because your review catches them. The machine runs overnight. Your own time is 90 minutes checking the picks and 40 minutes scanning the rejects in a grid, so 130 minutes against 7.0 hours by hand: about 4.8 hours back. The rejects scan is the part that earns the trust. A culling tool you never double-check is a gamble on every gallery. The cost side of paying for culling is in the tutorial on whether AI culling is worth it for wedding photographers.
Editing: what style learning needs and how to test consistency
Style learning means the tool is trained on your own edited images, then applies that look to new galleries. Vendors differ on what they ask for, and the differences decide who can use a tool this month.
| Tool | How it learns your look | Formats and where edits land | Where it processes |
|---|---|---|---|
| Aftershoot Edit | Import your edited photos, or a Lightroom preset plus a few edited samples; pre-built styles from 30-plus creators. No image count stated | RAW, JPEG, TIFF; one-click export to Lightroom, Lightroom Classic, Capture One, Photoshop, Bridge, Photo Mechanic | On your computer, per Aftershoot |
| Narrative AI Presets | Personal presets from at least 1,500 of your edited images; Artist presets from established photographers need no training | CR2, CR3, NEF, ARW, RAF, DNG and others; one-click import into Lightroom Classic on Mac or Windows; Lightroom, Photoshop and Capture One on Mac | Applying presets works offline; training needs a connection |
| Imagen Personal AI Profile | At least 2,000 edited photos per its help centre, all edited with one colour profile | Lightroom Classic, Lightroom, or Photoshop and Bridge; trains on your original edited files, not exported JPEGs or TIFFs; one profile per photo format and per colour or black-and-white style | Training photos are uploaded as DNG files and the profile trains after upload; ask how long uploads are kept |
| FilterPixel Style DNA | Trained on your edited Lightroom catalogue; Expert Profiles need no training | A long RAW list; sends to Lightroom, Capture One or Photo Mechanic | Originals stay local; only the embedded preview is uploaded, for its culling |
| Lightroom Classic | No training; you edit, with AI help from masking, Adaptive Presets and Content-Aware Remove | Your existing catalogue | Not stated on the Adobe pages I read; check before relying on it |
Count your catalogue before you compare anything. Under Imagen's rules, colour and black-and-white work need separate profiles, as does each photo format, and the training photos should share one colour profile, so each profile needs its own pile of edited frames. A photographer two years in with a few thousand edited frames may clear one threshold and miss another. Train only on edits that still represent your style, and keep your original files with their edits, because Imagen tells you not to train on exported JPEGs or TIFFs.
The consistency test is where editing tools earn or lose their place. Take 100 frames from each of three lighting scenes in your test gallery, apply the profile, and mark every frame as accepted untouched or needing a fix.
| Scene | Frames | Accepted untouched | Needed a fix | Fix rate |
|---|---|---|---|---|
| Ceremony, bright windows | 100 | 92 | 8 | 8% |
| Reception, mixed tungsten | 100 | 64 | 36 | 36% |
| Dusk portraits | 100 | 88 | 12 | 12% |
| All | 300 | 244 | 56 | 18.7% |
If that sample mirrors the 650-frame gallery, about 121 frames need a fix. At 1.5 minutes each that is 3.0 hours, plus roughly 1.8 hours to scan all 650 frames at ten seconds apiece. Editing falls from 9.5 hours to 4.8, so 4.7 hours back. The reception row is the warning: a 36% fix rate in mixed light means you need a separate preset for that lighting, or a manual pass on those frames, before you trust the tool unsupervised. Check skin tone too, by comparing the same person in church and at the reception, because a drift of that kind is what a couple notices. How to keep your look through the process is covered in the tutorial on cutting editing time while keeping your style.
Retouching: change as little as the client would accept
Retouching is the stage where an AI mistake is personal, so it earns the strictest test. Aftershoot's Retouch handles skin, stray hair, glasses glare and backgrounds, with edits synced across a batch. Evoto's wedding page lists skin smoothing, make-up, facial-feature and body adjustments with batch processing, plus an AI colour-matching feature that copies the look of a reference image across a series, and says it reads RAW as well as JPEG, PNG and TIFF. Its help centre says most features upload images to its cloud servers and delete them as soon as processing finishes, while cloud sync and colour matching store images only with your consent. Photoshop has Generative Fill, Generative Expand and an improved Remove tool built in, and Adobe ties its generative features to monthly credits that come with the plan, so read how many your plan includes.
The arithmetic explains the ranking. At four minutes a portrait, 45 hero frames cost 3.0 hours by hand. Even if an AI pass plus a two-minute review brought that to 1.5 hours, the most you could win is 1.5 hours a wedding, set against the stage where a mistake is most personal. So in the worked example retouching stays manual for now, and AI is tried only on a sample of 20 hero portraits, checked at 100% zoom with the original beside it.
Two realistic failures show what that check is for. Stray-hair removal can cut through the lace edge of a veil, leaving a clean line that looks wrong at print size. Skin smoothing can erase a scar or a laugh line the person regards as part of their face. Neither shows at thumbnail size, and both show up the moment a family member enlarges the photo. The safer routine is to retouch three representative portraits, send them to the client with the originals beside them, and batch only after they approve.
Generative tools raise a separate question: they add or replace pixels the camera never captured, such as extending a background or removing a guest. If your contract promises documentary coverage, or the client wants the images for editorial use, say so plainly. Wording for the contract and the gallery page is in the tutorial on AI disclosure for photographers.
RAW support, sidecars and where the files travel
Four questions about files decide whether a tool fits your jobs, and none appears in a feature list.
- Does it read your RAW format? Narrative lists CR2, CR3, NEF, ARW, RAF and DNG, FilterPixel lists a longer set that includes 3FR, ORF, RW2 and PEF, and Aftershoot says it handles RAW, JPEG and TIFF. Test files from every body you use, newest included, before the tool touches a paid job.
- Do edits and ratings go into sidecars? Aftershoot writes ratings for RAW files to XMP sidecars and says it does not overwrite originals. Imagen learns from the Lightroom edits attached to your originals, not from exported files. Confirm that your originals survive untouched.
- Where is the processing? Aftershoot says images are processed on your own computer. FilterPixel keeps originals local but uploads the small preview inside each RAW for DeepCull. Evoto uploads most images to its servers, and Imagen uploads training photos as DNG files. Narrative applies presets offline but needs a connection to train.
- Does the result return to your catalogue? One-click hand-off to Lightroom Classic matters if your delivery pipeline starts there. Windows users should note that Narrative lists its Lightroom, Photoshop and Capture One hand-offs for Mac only; its Lightroom Classic import works on both.
Wedding galleries contain children and guests who never signed a contract with you. The more of the pipeline that stays on your own machine, the shorter your list of questions to a vendor. Read each vendor's plan page for what is bundled (culling, editing, retouching), how images are counted, and how many devices or profiles one plan allows. Those details change more often than the features.
A five-step test on one finished wedding
You do not need a month-long trial, only one finished gallery you know well.
- Choose a wedding with variety: a bright ceremony, a mixed-light reception and a dusk portrait session. Your final selection and delivered edits are the answer key.
- Copy the originals, RAW files and sidecars together, into a test folder. Never run a test on your only copy.
- Cull test: run the tool in rate-only mode and record its picks, the overlap with yours, the misses and the minutes you spend reviewing.
- Edit test: apply the profile to 100 frames from each lighting scene, mark each frame accept or fix, and time the fixes.
- Retouch test, only if your contract allows it: 20 hero portraits at 100% zoom with the originals beside them.
Then let an assistant do the sums, and check two of them by hand:
Here is my time log for one wedding. Hours by hand: import 1.0, cull 7.0, edit 9.5, retouch 3.0, export 1.5. Culling test: 4,200 frames shot, 650 kept by me, 720 picked by the tool, 585 in common, 3 missed shot-list moments, 130 minutes of review. Editing test: 300 sampled frames, 56 needed fixes at 1.5 minutes each, and I would scan 650 frames at 10 seconds each.
Calculate recall and precision for the cull, the hours saved per stage, and the new total hours per wedding. Say which stages I should adopt AI for first and why, and flag any result where my sample is too small to rely on. Show every calculation.A trimmed answer might read like this (illustrative):
Cull: recall 585 / 650 = 90.0%; precision 585 / 720 = 81.3%. Review takes 130 minutes (2.2 hours) against 7.0 by hand, saving 4.8 hours. Edit: 56 / 300 = 18.7% need fixes, so about 121 of 650 frames at 1.5 minutes = 3.0 hours, plus 650 x 10 seconds = 1.8 hours, total 4.8 hours, saving 4.7 hours. New total: 12.5 hours per wedding. Adopt AI culling first, then editing.
The sums are right, and two things still need fixing before you act on them. In this illustration the answer skipped the warning you asked for: 300 frames from one wedding is a small sample, and the reception scene alone ran at 36%, so the blended 18.7% hides the scene that matters most. It also reported 90% recall without the three missed shot-list moments, which are the reason the cull passes for assisted use and fails for unattended use. Ask it to restate both points before you file the result.
A single wedding gives a first reading, not a verdict. Repeat the test on a second gallery of a different type, such as a winter wedding shot mostly indoors, before you commit to an annual plan.
What an hour back is worth, and what to promise clients
In the example, culling saves 4.8 hours and editing 4.7, so a wedding drops from 22.0 hours to 12.5. At an internal value of $35 an hour, 9.5 hours is $332.50 a wedding, or about $1,000 a month at three weddings. Divide any plan's monthly cost by $105 (your $35 times three weddings) and you have the hours per wedding the tool must save to pay for itself. Set that against the 9.5 hours you measured. If the break-even figure is a small fraction of 9.5, you have found your first purchase; if it is close to 9.5, the tool is marginal. Compare the same hours with outsourcing in the tutorial on AI culling versus outsourced editing.
Spend the saved hours deliberately. An 80-hour monthly editing budget fits 3.6 weddings at 22 hours each and 6.4 at 12.5. Most photographers want neither more weddings nor idle hours, so decide in advance how much goes to shorter delivery and how much to free weekends. A shorter promised turnaround is only worth advertising after your last three galleries have met it.
The cheapest new promise is a sneak peek within 48 hours. The tool culls the first 400 frames overnight, you review the picks in 25 minutes and edit 12 hero frames at about three minutes each. That is roughly an hour of your time, and the couple sees images before the honeymoon is over. It is also the safest place to try an editing profile in public, because the sample is small and you can fix every frame by hand.
Checks before an AI-assisted gallery goes to the couple
- Scroll the gallery in delivery order. Watch skin tone and white balance at each change of scene, not frame by frame. Drift shows up in sequence.
- Zoom into the key faces. Check blinks and expressions on the 30 or so frames the couple will print. A cull that preferred a sharper frame with half-closed eyes is a classic miss.
- Open every removal at 100%. Anything touched by Content-Aware Remove, Generative Fill or an AI retouch gets a close look at the edges.
- Match the count to the contract. If you promised up to 650 images, the final gallery should not quietly grow to 800 because a tool picked generously.
- Record what you used. Note the tool, the profile name and the date against the wedding. Profiles are retrained over time, and a record lets you reproduce the look for an album reprint or a second gallery.
- Check the wording the couple signed. Your contract and gallery page should say which stages use AI assistance and that you review every delivered image. A contract clause for edited-image ownership is in the tutorial on photography copyright and AI.
Choosing the first stage from your own timesheet
The worked example put culling first, but the timesheet is yours to read. This table turns the reasoning into a starting rule:
| If your paid jobs look like this | Start with | Why |
|---|---|---|
| 3,000 or more frames shot per job and weeks to deliver | Culling | No training needed, and your review catches the misses |
| A recognisable look and a large back catalogue of edits | Style-trained editing | The setup material already exists, and savings grow with frame count |
| Headshots or branding portraits where skin is the product | Retouching, on a sample first | It is the largest block of your time, but needs client approval |
| Same-night or next-day delivery from events | Culling plus a generic batch edit | Speed matters more than a personal style |
| Fewer than about 1,000 frames per job | Probably none yet | The hours returned are small, so fix import and delivery first |
Whatever the table says, run one gallery through the five-step test before paying for a year of anything, and keep the stage you do not trust in your own hands until a test has earned it its place. The tool that wins the cull test may be a poor retoucher, and a package that covers all three stages should still pass each test on its own.
Questions photographers ask before choosing an AI editor
Can I cull in one tool and edit in another?
Yes, and many photographers do. Pass the picks across as star ratings or colour labels through Lightroom Classic or XMP sidecars, then let the editing tool work only on the rated frames. Let one tool write ratings at a time, because two tools rating the same files can overwrite each other, and test the hand-off on a copied folder first.
What happens to my edits if I cancel the subscription?
Edits already written back into Lightroom Classic or XMP sidecars stay with your files, so delivered galleries are safe. What you lose is the trained profile, which usually lives in the vendor's software or account, so moving to another tool means training again from your edited catalogue. Keep your original edited files organised by year and style so that retraining stays possible.
Should a second shooter's frames go through the same profile?
Only after a test. A profile learns from the cameras and colour profiles in its training set, so a different body can shift skin tone and white balance. Run 100 of the second shooter's frames through it, compare matching moments side by side, and set up a separate profile if the fix rate is clearly higher than on your own frames.
Further reads
- Wedding Photographer Timeline Template: A Day Schedule Built With AI — A day schedule that protects the shots you must cull for
- Photographer Gallery Delivery: Reminders and Print Upsells With AI — What to send once the gallery is ready
- How to Price Photography Packages: A Cost-per-Shoot Model With AI — Put your editing hours into a cost-per-shoot price
- Photography Contract Template: Let AI Draft It, Then Check Rights — Contract clauses to review when AI touches client images
- ChatGPT for Photographers: What to Hand Over and What to Keep — What a photographer can hand to ChatGPT and what to keep
- How Can AI Help Photographers With Alt Text, Filenames and Captions? — Finish the gallery with alt text and filenames
- How to Use AI in a Photography Business: Start With Admin, Not Editing — The case for starting AI on admin before editing
- ChatGPT Prompts for Wedding Photographers: Timelines, Groups, Albums — Prompts for timelines, group lists and album planning
- Pixieset or Pic-Time for Photographers? Plans, Fees and AI Compared — Plans, print-sale commission, storage and AI features for two gallery platforms, with a yearly cost sum and a test to run on your own wedding photos.
- GDPR for Photographers: Client Galleries, Contracts and AI Tools — A photographer's data-protection checklist: gallery privacy settings, contract wording, retention periods, children's images and AI editing tools.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: vendor pages read in October 2026 for Aftershoot, Narrative, FilterPixel and Evoto (including Evoto's help centre on account settings and security); Imagen AI help centre (Create a Personal AI Profile with Adobe Lightroom Classic); Adobe Lightroom Classic and Photoshop product pages; Adobe help page on Assisted Culling in Lightroom Classic. The wedding gallery, timings and counts in the examples are illustrative.