alexpopov/photo-exif

Photo EXIF backfill

Alp shoots film, sends rolls to a lab, and receives JPEGs scanned by a Noritsu Koki HS-1800. The scanner writes its own identity into EXIF (Make=NORITSU KOKI, Model=EZ Controller) and leaves DateTimeOriginal / CreateDate empty. Before these import into Photos.app cleanly, EXIF needs backfilling.

The script

set-exif.sh wraps exiftool with three batched concerns:

1. Camera + lens — looked up by key from cameras.json
2. Sequenced date — base DateTimeOriginal plus N-second increments so Photos.app preserves frame order
3. GPS — decimal coords

Example:

~/dots/config/claude/skills/photo-exif/set-exif.sh \
  --camera canon-7 --lens canon-50-1.8 \
  --date "2025-07-06 12:00:00" --step 1 --tz "+02:00" \
  --gps "47.2913799,8.5277585" \
  /path/to/000071040001.jpg /path/to/000071040002.jpg ...

Always pass files in the order they should sort. For lab scans the filename order is already the frame order, so a shell glob works:

set-exif.sh ... "$DIR"/000071040{0001..0024}.jpg

Default to --no-backup. Alp doesn't want <file>_original files cluttering the scan dirs. Omit it only if there's specific reason to keep a backup for that run. To clean up backups left by an earlier run, move them to /tmp/ (don't delete outright — Alp prefers a recoverable step).

Use --dry-run to preview the exiftool commands before running.

Private location presets (locations.local.json)

For locations that shouldn't be in the public repo (home, friends' places, etc.), use --location KEY instead of --gps. The script reads locations.local.json from the skill dir; that filename is gitignored (config/claude/skills/photo-exif/*.local.json).

Schema:

{
  "locations": {
    "home": { "description": "Home", "lat": "47.xxx", "lon": "8.xxx" }
  }
}

When the user names a private location, add it to locations.local.json (which already exists if they've used the feature before). Do not put private coords in cameras.json, the SKILL.md, or anywhere else under version control — that includes commit messages and PR descriptions.

Presets (cameras.json)

Two flat dictionaries: cameras and lenses, keyed by short slugs. Each entry has a human description and an exif map that becomes -Tag=Value args.

When the user mentions a camera or lens that isn't in cameras.json yet, add it:

• Camera: at minimum Make, Model.
• Lens: LensMake, LensModel, FocalLength, FocalLengthIn35mmFormat, MaxApertureValue (lens's widest aperture — not per-shot FNumber, which we don't know for film).

Confirm the slug and description with the user before saving the preset.

Resolving locations

The user can give a location in any form — proceed without asking them to convert:

• Google Maps share URL or short link (maps.app.goo.gl/..., google.com/maps/..., goo.gl/maps/...): follow the redirect and pull coords out of the resolved URL:

  curl -sIL "<url>" | grep -i '^location:'
  

  The resolved URL has both @LAT,LON,ZOOMz (map-view center) and !3dLAT!4dLON (the actual pin). Prefer !3d!4d — for pinned places those can differ from @ by 100m+. If only @ is present (a "dropped pin" with no place data), use that.
• Place name (e.g. "Wildnispark Langenberg, Zürich"): the user expects you to figure it out. Ask only if genuinely ambiguous.
• Decimal coords (e.g. 47.38, 8.54): use directly with --gps.

For private locations, save under a key in locations.local.json instead of passing --gps (see above).

Workflow checklist

1. Read the source directory and current EXIF on one file (exiftool <file>) — confirm what's missing.
2. Gather from the user: camera, lens, date(s), location. If they give a range of files belonging to one shoot, note where the next shoot starts.
3. If a new camera/lens, add to cameras.json and confirm slug.
4. Run with --dry-run first if anything's uncertain.
5. Apply.
6. Re-read EXIF on one file to verify (exiftool -DateTimeOriginal -Model -LensModel -GPSPosition <file>).

Exporting to HEIC for Photos.app import

Lab/home TIF scans are 100+ MB each (16-bit, uncompressed). For Photos.app import they should be downscaled + compressed first so the iCloud Photos library doesn't bloat. The canonical workflow:

1. Tag the TIFs (this skill) — date, camera, lens, GPS, film stock. The TIFs are the canonical archive; keep them in iCloud Drive (~/Library/Mobile Documents/com~apple~CloudDocs/Pictures/Imported/...).
2. Convert to HEIC with tif-to-heic.sh — defaults to 3840 long-edge max, q=80, output to a separate dir (typically ~/Desktop/Scans/... or similar).
3. Import the HEICs into Photos.app (open -a Photos <dir>/*.heic).

Why a wrapper script

sips is fast (hardware HEIC encode on Apple Silicon) and preserves most EXIF — except Keywords and XMP Subject are silently dropped during TIF→HEIC. So film-stock tags vanish unless you re-copy them. tif-to-heic.sh does the sips encode then runs exiftool -tagsfromfile to put Keywords/Subject back.

Usage

~/dots/config/claude/skills/photo-exif/tif-to-heic.sh \
  --out "$HOME/Desktop/Scans/Canon 7/Porta 1" \
  "/Users/alp/Library/Mobile Documents/com~apple~CloudDocs/Pictures/Imported/Canon 7/Porta 1"/*.tif

Defaults: 3840 max long edge, quality 80, 8 parallel workers. For 34 frames expect ~50–60 seconds and ~50 MB output.

Quality reference (from earlier benchmarks on a 5034×3437, 108 MB TIF):

• q=95 @ 3840: ~5.5 MB (19× compression, indistinguishable at viewing size)
• q=80 @ 3840: ~1.5 MB (75× compression, fine for tag/sort, mild grain softening if you pixel-peep)

Timezones

DateTimeOriginal is a wall-clock string with no zone. Photos.app uses the file's location to infer the zone, but setting OffsetTimeOriginal explicitly avoids surprises. Common values for Alp:

• Zürich summer (CEST): +02:00
• Zürich winter (CET): +01:00