Photographer Batch Workflow for AI Metadata

A two-week trip easily produces 1,000 keepers, sometimes a lot more. The shooting was the fun part. The metadata work that follows is what most photographers procrastinate on, and the longer you put it off the harder it gets, because by month four you cannot remember which lake was which.

A clean batch workflow turns the metadata pass from a multi-day ordeal into a couple of evenings. The trick is sequencing. Do the human work where humans are needed, do the AI work where AI is faster, and do not duplicate either step.

This guide walks through a generalized 1,000-image trip workflow. It covers when to sort before AI vs after, when ranking actually helps, how to handle near-duplicate burst shots, how to round-trip back into Lightroom via ZIP export, and how to think about credits vs subscription so you do not overpay.

The 1,000-Image Trip Workflow at a Glance

The full pass, end to end, looks like this. Ingest and back up. First-pass cull in Lightroom or Photo Mechanic. Group by location or day. Run AI metadata in batches. Review the captions. Round-trip back into Lightroom. Optional second pass for ranking and final selects.

The order matters. The most common mistake is running AI metadata on the entire raw take before culling. You waste credits on photos you will never publish, and the metadata pass takes much longer than it needs to.

Sort Before vs After AI

The general rule is sort before AI when you have an obvious throwaway pile, and sort after AI when you cannot tell what is what until it is described.

For most current trips, the sort-before approach is right. For most archive cleanup projects, the sort-after approach is right.

When to Use Ranking

Ranking is the feature that asks the AI to score and order images by quality, by interest, or by subject relevance. It is genuinely useful in a couple of specific cases and a waste of credits in most others.

A common pattern is to run a fast ranking pass on the unculled set first, use the rankings to pick the top 200, then run the full caption and keyword pass on those 200. This costs less and produces better captions because the AI is spending more time on each image that survives.

Handling Near-Duplicates

Near-duplicates are the hidden cost in any large batch. Burst shots, bracketed exposures, panorama frames, and "let me try one more" frames all produce visually similar files. Captioning each one separately is wasteful and produces a catalog full of nearly identical descriptions.

The cleanest approach is to collapse near-duplicates before the AI pass. Pick the best frame from each cluster, hide or reject the rest, and only send the keepers. If the rejects matter for later decisions, keep them but exclude them from metadata work.

For panoramas and HDR brackets, only the merged final needs full metadata. The source frames can inherit a brief auto-caption like "HDR source frame" or "panorama tile" so they are still findable but do not consume the full descriptive treatment.

If you do run AI on near-duplicates, expect the captions to be similar. That is correct. The captions describe the image and the images are similar. Use ranking output to pick the keeper rather than reading every caption to find subtle differences.

ZIP Export to Lightroom

Once the AI batch is done, the round trip to Lightroom is the part that turns AI output into a permanent part of your catalog. The Lightroom-mode ZIP export packages JPEGs with captions, headlines, and keywords baked into the IPTC headers, ready for re-import.

The mechanics. In Lightroom, export the selects as JPEGs at small or medium size with original filenames. Drop the folder into PhotoScanr. Run the batch. Download the ZIP. Unzip it somewhere convenient. Back in Lightroom, select the original raw or full-res files, then "Metadata > Read Metadata from Files." Lightroom matches by filename and pulls the IPTC data into the catalog and into the XMP sidecar.

A few caveats. Lightroom will warn before overwriting catalog metadata with file metadata. Read the warning. If you have made any catalog edits to keywords or captions since exporting, those will be replaced. For a fresh AI pass on previously untagged photos this is what you want. For a second pass on already-tagged photos, sync more carefully.

Credits vs Subscription

The pricing question that comes up most often is whether to subscribe or to buy credits. The honest answer depends on your volume and how spiky it is.

The current credit packs are 1,000 credits for $20, 5,000 for $90, and 10,000 for $160. The per-credit price drops at higher tiers, so for a known large project the 5K or 10K pack is the better deal. For a one-off small project, the 1K pack is fine.

A common combined pattern is a Pro subscription for the steady weekly work, plus a credit pack each spring before the big trip season. That keeps the monthly cost low and absorbs the seasonal spikes without an upgrade cycle.

Putting It All Together

A 1,000-image trip does not have to take a month to process. Cull first. Group by day or location. Run a ranking pass to identify keepers if the set is large or unfamiliar. Run the full AI pass on the keepers with style preferences and grounding set. Round-trip the ZIP back into Lightroom. Done.

The workflow scales up to 5,000 image trips and down to 200 image weekend shoots without changing shape. The features stay the same. Only the batch sizes and credit consumption shift.

For specific Lightroom integration details, see the AI photo metadata for Lightroom 2026 guide. If you also do client deliverables on the side, the event captions guide covers ZIP plus CSV handoffs that work with the same batch tools.