Contact
Speranda/Lightroom AI Bridge/Docs

Editing and culling workflows

The everyday jobs Lightroom AI Bridge is built for, and how to get the best out of each. If you have not installed it yet, start with the User Guide.

The pro edit: one photo from RAW to finished

Select one photo in Lightroom and type /lightroom-bridge-pro in Claude Code. The agent works as a senior photo editor would, and asks you nothing:

  1. The brief. Before touching a slider it writes down what the photo is about, where the eye goes, what is competing for attention, the defects, and one sentence on how the finished photo should feel. That sentence becomes the test for done.
  2. Snapshot. A restore point, so everything that follows can be undone in one step.
  3. Foundation and geometry. Profile and lens corrections first, then level and crop, because tone judged on pixels that will be cropped away is judged wrong.
  4. Global tone, then color. White balance, exposure, highlights and shadows, white and black points, contrast, then presence and a disciplined color palette.
  5. Cleanup, then masks. Dust and distractions go before any AI mask exists. Then sky, subject and background masks shape where the eye goes: lift the subject, calm the competition.
  6. The grade and detail. Color grading on top of the finished base, then noise reduction and sharpening at real scale.
  7. The quality check. A final render, audited as if someone else made it: clipping, skin and sky colors, halos, edges, and whether the brief's sentence came true. If the edit reads as edited, the strongest move is dialed back.

Add a sentence to steer it: /lightroom-bridge-pro warm and quiet, keep the fog. The pro edit never exports and never changes ratings or flags.

Quick edits in plain words

For a single change, just ask: "open up the shadows", "the sky is too cyan", "straighten this". The agent renders before and after so you can compare, and every step lands in History.

Batch edit a set of photos

Select the set in Lightroom and ask: "edit the first one properly, then apply the same recipe to the rest."

The agent finishes one photo, turns the edit into a recipe, and applies it photo by photo, reading each one back to confirm the values landed. When the light changes across the set, ask it to group the photos by light first: thirty frames from one golden-hour walk are one recipe with small per-photo adjustments, not thirty separate edits.

Save a look as a Lightroom preset

"Save this look as a preset" writes a real Develop preset into Lightroom, in a group called Claude. It appears in the Presets panel after Lightroom restarts, and from then on you and the agent can both apply it by name. Crop, masks and retouching are left out on purpose: a preset is a look, not a composition.

Cull a shoot

The technical cull: Lightroom does it best

Focus misses, closed eyes and near-identical burst frames are best caught by Lightroom's own Assisted Culling (Lightroom Classic 15.4 or later), which works at full resolution. Judging fine sharpness is the weakest thing AI vision does, so run this first.

Small folders: contact sheets

  1. Open the folder in Lightroom and press Cmd+A to select everything. The bridge only sees selected photos.
  2. Ask: "review this folder and flag the best shots."

The agent renders every photo without disturbing your selection, lays them out on labeled contact sheets, reads them, and flags its picks. It explains each pick and points out clusters of near-duplicates so you can keep one per cluster.

Hundreds or thousands of frames: the two-stage funnel

For big shoots, rating happens in two passes, because image size matters more than the model:

  1. Triage on contact sheets. Every photo, small, judged quickly: obvious rejects go, bursts collapse to their best frame, the rest get rough stars.
  2. The top tier at full size. Only the survivors, one large image each, judged in small side-by-side comparisons. This is the pass that separates four stars from five, and it rescues the night and blue-hour frames that small thumbnails make look flat.

Ratings are written to each photo individually and read back, so a star never lands on the wrong frame. Treat AI ratings as advice: do a final check at 100 percent on your shortlist before you publish.

Measure instead of arguing

Some questions have numbers for answers. Ask "measure the tilt" or "does the subject win?" and the agent uses the measuring tools in the project:

  • Tilt and keystone across the whole frame.
  • Plumb from one structure, such as a tower or a doorframe, for when foliage or water lines fool the whole-frame reading.
  • Attention check, which compares the subject against every other large area for brightness, saturation and contrast.

When your eye and the whole-frame number disagree, measure the thing your eye is looking at.

Get help