Lightroom Classic: AI analysis progress (Assist Culling / Face Detection) resets after a clean quit + relaunch, once it reaches 100% — reproducible on both macOS and Windows
Description of Issue
After Assist Culling (and Face Detection) analysis for a catalog reaches
100% complete, quitting Lightroom Classic normally (File > Quit, not a
crash/force-quit) and relaunching causes the "remaining photos to
analyze" counter to jump back up — i.e. photos that were already
analyzed are treated as unanalyzed again. This is reproducible across
two different machines (Windows and macOS) and multiple catalogs,
including catalogs rebuilt from scratch via "Export as Catalog".
Lightroom Classic Version Number (Help > System Info)
Lightroom Classic version: 15.3.1 [202605261338-e36f8566]
License: Creative Cloud
OS Version Number
- Machine A (macOS): Mac OS 26, version 26.5.2 [25F84], arm64
- Machine B (Windows): Windows (native, catalog originally migrated
from Mac to this Windows machine)
Camera Model
N/A (issue is not camera/file-specific; reproduced on catalogs with
56,000+ mixed JPEG/RAW files)
Step-by-step Reproduction Instructions
1. Open a catalog where Assist Culling / Face Detection is actively
analyzing photos (Library module, background task indicator top
left shows "Analyzing photos").
2. Let the analysis run until it reaches 100% (all background tasks —
Assist Culling, Face Detection, Duplicate Detection — show 0
remaining / "Completed").
3. Quit Lightroom Classic normally via the application menu
(Lightroom Classic > Quit Lightroom Classic). Confirm it is a clean
exit (no crash dialog).
4. Relaunch Lightroom Classic. The same catalog reopens automatically.
5. Open the background task progress popover (click the small progress
indicator near the top-left "analyzing" icon) and check the
remaining count for Assist Culling / Face Detection.
The Expected Result
The remaining count should stay at 0 (still 100% complete), since all
photos were already fully analyzed before the clean quit.
The Actual Result
The remaining count jumps back up to a large non-zero number, meaning
previously-completed analysis is discarded and has to be redone. This
was reproduced with the following measured data:
Machine B (Windows), catalog "Lightroom Catalog.lrcat" (56,109 photos):
- End of session 1 (clean quit): 51,819 / 56,109 analyzed
- Immediately after relaunch (session 2): regressed to 37,985 / 56,109
- Ran to 100% again in session 2, clean quit
- Immediately after relaunch (session 3): regressed again to 37,959 / 56,109
- Rebuilt the catalog from scratch via File > Export as Catalog (new
.lrcat file, new location on disk) — same regression pattern
reproduced: reached 56,109/56,109, clean quit, relaunch regressed to
38,160/56,109.
Machine A (macOS), catalog "Lightroom Catalog-A-v13.lrcat" (56,929
photos, a different/independent catalog from the one above):
- Assist Culling, Address Lookup, and Face Detection all reached 100%
complete (0 remaining) during the session.
- Quit Lightroom Classic normally (Lightroom Classic > Quit Lightroom
Classic). Confirmed clean exit (returned to desktop, no crash
dialog).
- Relaunched. Same catalog and same selected photo restored
automatically (state was otherwise preserved correctly).
- Checked background task popover immediately after relaunch:
- Assist Culling: regressed from 0 remaining to 11,425 remaining
(then continued decreasing again: 11,272 -> ...)
- Face Detection: regressed from 0 remaining to 3,198 remaining
(then continued decreasing again: 3,096 -> ...)
Note: on the same Mac, an earlier test on a *different* catalog
("Lightroom Catalog-A-v14.lrcat", 56,109 photos) that had NOT yet
reached 100% (only ~5% analyzed) did NOT show a regression after a
clean quit/relaunch — the remaining count kept decreasing normally
across the restart. This suggests the regression is specifically
triggered by reaching 100% completion before quitting, not by the
quit/relaunch cycle itself, and is not OS-specific (reproduced on both
Windows and macOS).
Troubleshooting / Workarounds Already Tried (none resolved it)
- Removed read-only attributes recursively on the catalog folder
(Windows: attrib -R -H -S)
- Granted full NTFS control permissions to the user account (Windows:
icacls /grant ... /T), verified via Security tab with no explicit
deny entries
- Confirmed the catalog folder is not inside a synced folder (OneDrive
/ Dropbox / iCloud / Synology Drive all excluded)
- Confirmed Windows "Controlled Folder Access" (ransomware protection)
is off
- Ran File > Optimize Catalog (completes without error)
- Renamed/rebuilt the external AI stores (.lrcat-data and the
*.lrdata Helper/Previews/Sync folders) to force Lightroom to
recreate them from scratch — regression still occurred
- Rebuilt the entire catalog via File > Export as Catalog (fresh
.lrcat file, new location on disk, ratings/flags/collections
correctly carried over) — regression still occurred
- Checked %LOCALAPPDATA%\Adobe\Lightroom (Caches/Logs/Preferences) and
%LOCALAPPDATA%\Adobe\ACPLocal.Next\CommonDB\1\acplnext-common.db on
Windows — no evidence these are involved (the .db file's mtime never
changed even after a full analysis pass completed)
Given the same completed-state regression reproduces on a second,
independent catalog on a second machine (macOS) with none of the above
mitigations applied differently, this looks like a bug in how
Lightroom Classic persists/validates the AI analysis progress
watermark for a catalog when it reaches 100%, rather than a
permissions, sync, or OS-specific issue.
Happy to provide the affected .lrcat files, screen recordings, or
further reproduction detail if useful.
