Train an AI Editing Profile from Your Lightroom Catalog

This page is about preparation and thresholds. For which tool to buy, see the best AI wedding editing software; for a two-tool comparison of how Aftershoot and Imagen learn, see Aftershoot Edit vs Imagen.

What the vendors publish

Published minimums vary, and one important vendor gives no number. The table shows only what the pages read state.

As of 8 November 2026. "Not found in sources reviewed" means we looked and did not see it.

Minimums are what the tool asks for to start, not evidence that a profile built at that size will match your look. Nobody publishes a fit-versus-size result. The Imagen figure comes from its plugin-page FAQ; its support article is source inaccessible (403). See the Imagen review for its training path. Others: FilterPixel, Neurapix.

Prepare the catalogue before you train

A profile learns from whatever you give it, so curation matters more than raw volume. These steps follow the guidance Aftershoot and Narrative publish (consistent edits, variety, separate profiles per style) and general Lightroom practice.

1. Choose images you are happy with. Narrative asks for images edited consistently; Aftershoot asks for consistent edits. Leave out early-career work in a style you have dropped.
2. Pick a Lightroom collection, not the whole catalogue. Build a collection from finished weddings only, so rejected or half-finished edits do not teach the tool. A flagged or star-rated subset is a simple way to mark the final selects.
3. Cover the range of conditions you deliver. Aftershoot's article asks for variety and enough images per lighting condition. For weddings that means getting-ready light, ceremony, flash receptions and outdoor portraits, not 3,000 near-duplicates of one venue.
4. Separate looks into separate profiles. Aftershoot says to use separate profiles for different styles, and Imagen recommends a profile per niche. A colour and a black-and-white look, or a bright and a moody look, do not belong in one training set. Narrative's Premium plan includes one colour and one black-and-white Personal AI Preset; Ultra includes five of each (Narrative pricing).
5. Make the edits finished. Training learns from edits as saved. If you routinely fix skin tone in Photoshop after Lightroom, that correction is not in the catalogue.
6. Count honestly. Confirm the selection meets the vendor minimum after you remove outliers, not before.

Tool-by-tool: what is documented

Narrative

Narrative scans your Lightroom catalogs, and requires an internet connection for training; its presets page says AI Preset edits need Lightroom Classic or Lightroom CC and names no Lightroom version for training, so Classic-only training is not confirmed. Its Select page says training a Personal AI Preset uploads edited images and edit data, while culling and assessing photos stays on your computer; a "Help improve Narrative" option is off by default. Narrative's presets page says you can then work offline for up to 30 days from your last sign-in. Personal presets start at the Premium tier. Detail: Narrative pricing explained.

Imagen

Imagen builds a Personal AI Profile from a Lightroom Classic catalog connected through its plugin and processes edits on its servers from uploaded smart previews. The step-by-step profile article is on its support site (source inaccessible), so those steps are not described here. Check the live steps in your account.

Aftershoot

Aftershoot Edit can be trained on your past edits or on presets and creator styles, per its pricing-tiers post, which states a 2,500-image minimum. The support article gives conditions rather than a count, so the two Aftershoot pages differ in what they say; confirm the number in your trial. The post also says custom profiles are deleted when Edit ends unless you pay $4 a month to keep them stored.

FilterPixel and Neurapix

Both describe learning from your Lightroom edits, but the count and steps were not found in sources reviewed, and their terms and privacy policies were not read. Data handling is vendor statements only: FilterPixel says originals never leave your computer and only the embedded RAW preview uploads (FilterPixel); Neurapix offers local editing and says you can keep RAWs on your own computer (Neurapix). Do not budget for either on the basis of this page.

Where your files go

Training is the step that uploads images, and the data terms conflict or could not be read, so no tool here is recommended on confidentiality grounds.

Narrative

per its Select page, training a Personal AI Preset uploads edited images and edit data and culling stays on your computer; sharing image previews to improve Narrative is optional and off by default; the presets page says only that training needs a connection. Its privacy policy (2022) excludes Visitor and Customer Data from its scope and refers to its Terms (source inaccessible, 404), so its silence on AI training says nothing about customer images. Profile fate on cancelling: not found in sources reviewed.

Imagen

smart previews upload; its privacy policy says content is used to teach your style and stored until you request deletion. Terms of Service: not obtained.

Aftershoot

per its transparency page, images are processed locally for batch culling, editing and retouching; custom editing-profile training uploads the images you choose; with learning on, culling sends small-resolution versions to its servers, which you can turn off at album import; storage is on Google Cloud, with deletion on request generally taking 7 to 14 days. The homepage separately says files never leave your machine and images are never stored on its servers, and the 2023 privacy policy says images and edits it collects are used solely for training its models; the policy says it applies to all information collected through its Websites and Service, and whether locally processed images count as "collected" is not stated. All vendor statements; terms of service source inaccessible (404).

What if your catalogue is too small

If you fall short of a vendor's minimum, the options are documented rather than clever. Narrative's 1,500 is the lowest stated minimum among the tools above with a published number (Neurapix's page refers to building a SmartPreset from 20 or 500 images without calling it a minimum). You can also start from the tools' prebuilt styles: Narrative lists Core and Marketplace AI Presets from its Standard tier, and Aftershoot's homepage mentions a marketplace of 30-plus pre-built styles. Those are covered on AI profiles vs Lightroom presets. Beyond that, the honest answer is to keep editing manually and revisit the training later.

After training: checks that matter

Run the new profile on a gallery that was not in the training set and compare it with how you would edit it. Look at white balance on mixed lighting, skin tones under coloured venue lights, and how the profile handles scenes with no equivalent in your catalogue. Then see how the batch runs in practice in batch editing wedding photos with AI.

Check this in your trial

1. Whether the tool enforces its stated minimum or proceeds with fewer images.
2. How long training takes, and whether it ties up your machine.
3. How the profile handles a venue absent from the training set.
4. Whether you can retrain or add images later, and whether old profiles are kept after cancelling.
5. What the tool says it uploaded, whether any learning upload can be switched off, and how to request deletion.
6. Whether Lightroom reads the returned edits and keeps them adjustable.

Original test pending: WeddingWorkflowLab plans to build profiles from catalogues of different sizes and curation on a licensed gallery and score each against the photographer's manual edits, to show fit by catalogue size.