iPhone Photo Backup for Windows

Back your iPhone photos up to a Windows PC and end up with a library you can actually navigate: one folder, every file named for the date it was taken, and no duplicates.

IMG_7533.HEIC   ->   20231008-001.heic
IMG_7534.HEIC   ->   20231008-002.heic
IMG_7534.MOV    ->   20231008-002.mov     (Live Photo, kept paired)

Sort by filename and you are sorting by time. Plug the phone in again next month and run it again; it only touches the new arrivals.

PowerShell 5.1+, no dependencies, no cloud, no account. Nothing is ever deleted — duplicates move to a quarantine folder for you to review — and every run is reversible.

Why this is needed

Copying photos off an iPhone gives you folders like 100APPLE full of IMG_0001.HEIC, and the problems start immediately:

• Filenames are meaningless and collide. iPhones recycle IMG_0001 endlessly, so the same name covers unrelated photos across folders and across imports.
• File timestamps are wrong. Copying sets them to the copy date, not the day you took the photo. Sorting by "date modified" sorts by when you plugged the cable in.
• You import the same photos repeatedly. Every import overlaps the last, and Windows will happily keep IMG_0001 (2).HEIC forever.
• Windows silently converts HEIC to JPEG on import unless you change a setting, so the same photo ends up stored twice in two formats that share no bytes.

This handles all four.

Getting started

Copy photos off your iPhone however you like — Windows' Import pictures and videos, dragging DCIM folders out of File Explorer, or an existing pile you already have. Drop them anywhere inside your Pictures folder. The layout does not matter; nested subfolders are fine.

# 1. Look, don't touch. Read-only: writes a plan, changes nothing.
.\Migrate-Photos.ps1

# 2. Read reports\run-<timestamp>\plan.csv -- every rename it intends to do.

# 3. Do it.
.\Migrate-Photos.ps1 -Execute -PlanPath ".\reports\run-<timestamp>\plan.csv"

# 4. Changed your mind? Put everything back exactly as it was.
.\Undo-Migration.ps1 -ManifestPath ".\reports\run-<timestamp>\manifest.csv"

Point it elsewhere with -Root "D:\Photos". Nothing is modified until step 3, and step 3 is fully reversible by step 4.

Every import after the first: add -Incremental so files already sorted are left alone. See Adding more photos later.

How dates are worked out

File timestamps are the last thing to trust, so real capture metadata is read first:

Priority	Source	Notes
1	EXIF Date taken	Photos. Trusted outright
2	Media created	Videos. Accepted only if it agrees with the folder's month
3	Date embedded in the filename	Screenshots, WhatsApp saves, PXL_20240229_...
4	LastWriteTime, if it agrees with the folder's month	Carries a real day-of-month
5	Month from a YYYYMM folder name	Day defaults to the 1st
6	LastWriteTime	Rough, but usually the right era
7	CreationTime	Last resort. Flagged as low-confidence
—	Unreliable	Everything failed. Flagged for manual review

Every row in plan.csv records which source won, in its DateSource column. Sort by it and spot-check anything below priority 3 before you execute.

Implausible dates are rejected rather than used — a video carrying an empty timestamp that reads as 1969 falls through to the next source instead of landing under 19691231-NNN.

How duplicates are handled

Deduplication is content-based (SHA-256), never name-based — which matters in both directions:

• Same bytes, different names → genuine duplicates. One copy is kept; the rest move to quarantine.
• Same name, different bytes → not duplicates. Because iPhones recycle IMG_0001, name-matching would silently destroy unrelated photos. Every distinct image survives. They are listed in same-name-different-photo.csv.

That guarantee is about images, not files. One name can cover three files that are only two distinct photos — two byte-identical, one different. The redundant copy is still quarantined; both real photos survive:

File	Image	Action
100APPLE\IMG_0213.JPG	A13988E3…	Keep
101APPLE\IMG_0213.JPG	A13988E3…	Quarantine
102APPLE\IMG_0213.JPG	0BBB0934…	Keep

Files kept always equals distinct images. Nothing visually unique is ever quarantined.

Where several copies are identical, the one with the most trustworthy date wins, then the shallowest path, then alphabetical order — so runs are repeatable.

.AAE sidecars are exempt from deduplication. Apple writes byte-identical edit sidecars for unrelated photos, so hashing them would throw away real edit data. They are matched to their photo by name instead.

Naming

YYYYMMDD-NNN.ext — 20250804-001.jpg, 20250804-002.jpg.

Year-first so a filename sort is chronological, and zero-padded to three digits so -002 does not sort after -010. NNN restarts each day and follows the order the shots were actually taken.

Live Photos stay paired. The HEIC and its MOV share a base name, so they get the same number and stay together:

20250804-007.heic
20250804-007.mov
20250804-007.aae

Use -NoPairGrouping to number every file separately.

Labels are preserved. Rename a file to 20250804-007-Birthday.heic and later runs leave it alone, so you can annotate favourites without breaking anything.

Adding more photos later

Copy the new photos in, then run with -Incremental. Files already carrying a YYYYMMDD-NNN name in the library root are treated as settled:

• Never renamed, renumbered or moved. They show as Skip in the plan.
• Numbering continues. A new photo from a day that already ends at -003 becomes -004.
• The copy you already have wins. Re-importing a photo you already filed quarantines the new copy, not the settled one.

# 1. Copy new photos into the library. Copy, don't move -- leave the
#    originals on the phone until you have checked the result.

# 2. Audit.
.\Migrate-Photos.ps1 -Incremental

# 3. Optional: find the same photo saved as both HEIC and JPEG.
.\Find-NearDuplicates.ps1 -PlanPath ".\reports\run-<timestamp>\plan.csv"

# 4. Check "Already migrated, untouched" matches your current file count,
#    then read plan.csv and duplicates.csv.

# 5. Apply.
.\Migrate-Photos.ps1 -Execute -PlanPath ".\reports\run-<timestamp>\plan.csv"

Always use -Incremental on a library you have already migrated. Without it, numbering is recalculated from scratch and thousands of already-sorted files get renamed for no reason.

Every run re-hashes the whole library to catch duplicates, so audits get slower as it grows — roughly a minute per 12 GB. Only the audit is affected; applying a plan is fast, because moving files within a drive is instant.

Quarantined duplicates need no special handling on later runs: quarantine is skipped during scanning, but every file in it is byte-identical to one still in the library, so an incoming copy still matches.

Same photo, HEIC and JPEG

Windows converts HEIC to JPEG on import unless you set Keep Originals. Import once each way — or on two PCs — and you hold both renditions of the same photo. They share no bytes, so they are not duplicates in any technical sense.

Both are kept, and that is the right default. HEIC is the original at about half the size; JPEG opens anywhere without conversion. Neither can be regenerated from the other, and the overlap is typically a fraction of a percent of the library — not worth the risk of throwing one away.

If you want to see them anyway:

.\Find-NearDuplicates.ps1 -PlanPath ".\reports\run-<timestamp>\plan.csv"

This writes possible-duplicates.csv listing each pair with both filenames, formats and sizes. It only reports — there is no option to act on it. If you decide you want only one format, move those files to the quarantine folder yourself using the filenames in the CSV.

Pairs are matched on the camera filename stem plus the capture instant together; either alone is unreliable, since iPhones recycle IMG_#### numbers and Windows exposes Date taken only to the minute. Live Photo video halves are excluded, and Apple's edited renditions (IMG_E7533) count as separate photos.

Once a file is renamed, nothing on disk remembers it arrived as IMG_7533.HEIC, so Migrate-Photos.ps1 records that in reports\library-index.csv as it goes. The script reads it automatically; there is nothing to pass or maintain.

What the reports tell you

reports\run-<timestamp>\

File	Contents
plan.csv	Every planned action: source, destination, new name, capture date, date source
duplicates.csv	Each quarantined copy paired with the twin kept in its place. Sorted largest first
same-name-different-photo.csv	Files sharing a name that are genuinely different photos
possible-duplicates.csv	From Find-NearDuplicates.ps1: the same photo in two formats
ignored-files.csv	Non-media files, left where they are
manifest.csv	Written by -Execute. Feeds Undo-Migration.ps1

reports\library-index.csv sits alongside them and accumulates across runs.

Before executing, three things are worth a look:

1. The low-confidence date count in the summary. If it is not zero, filter plan.csv by DateSource and spot-check.
2. A few rows of duplicates.csv. Matches are byte-identical, so both files are the same image and the choice only affects which folder it came from. Verify a pair independently with fc /b "<KeptPath>" "<DuplicatePath>".
3. The size of same-name-different-photo.csv, to confirm nothing collapsed that shouldn't have.

Safety

• Two passes. The scan cannot modify anything; execution only replays a CSV you have read.
• Nothing is deleted, ever. Duplicates are quarantined, not removed. Emptied folders are reported, not deleted — the script prints the cleanup command and leaves the decision to you.
• Nothing is overwritten. An occupied destination gets a _dup1 suffix and a warning.
• Fully reversible. Undo-Migration.ps1 restores original paths and recreates the original folder tree.
• Plans know their own library, so applying one against the wrong -Root warns and corrects itself rather than acting on the wrong folder.
• -CopyInstead copies instead of moving, leaving the source tree untouched. Needs room for a second copy.

Keep the reports\ folder. It holds the undo manifests and the camera-name index.

Parameters

Migrate-Photos.ps1

Parameter	Default	Purpose
-Root	%USERPROFILE%\Pictures	Library root. Scanned recursively; files land back here
-Execute	off	Apply a plan. Requires -PlanPath
-PlanPath	—	The reviewed plan.csv to apply
-Incremental	off	Adding to an already-migrated library. Leaves settled files untouched
-CopyInstead	off	Copy instead of move
-NoPairGrouping	off	Number Live Photo / sidecar sets separately
-QuarantineName	_DuplicatesQuarantine	Quarantine folder, created under -Root
-ReportRoot	.\reports	Where run artifacts go. Outside the library by default

Find-NearDuplicates.ps1

Parameter	Purpose
-PlanPath	The plan to analyse (required)
-IndexPath	Camera-name index. Found automatically beside the plan
-OutputPath	Report destination. Defaults to the plan's folder

Undo-Migration.ps1

Parameter	Purpose
-ManifestPath	The manifest to reverse (required). Supports -WhatIf

Supported files

Photos .jpg .jpeg .png .heic .heif .gif .bmp .webp .tif .tiff, raw .dng .raw .cr2 .cr3 .nef .arw .orf .rw2, video .mov .mp4 .m4v .avi .mkv .3gp .mpg .mpeg .wmv, sidecars .aae .xmp .thm.

Anything else is listed in ignored-files.csv and left exactly where it is.

Testing

Verified end-to-end on sandboxes seeded with real photos plus deliberately planted exact duplicates and name-collisions-with-different-content:

• Flatten, rename and quarantine produced exactly the planted duplicates and no others
• Unique SHA-256 count identical before and after — no unique image lost
• Undo restored every file to its original path and folder structure
• Incremental mode left settled files untouched, continued numbering correctly, and made settled copies win duplicate contests
• The folder-date parser is unit-tested against folder names that must parse and names that must not

Notes

• Windows only: reads photo metadata through Shell.Application COM.
• If script execution is blocked: Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass
• reports\ is gitignored. It contains the paths and filenames of your library.

A note on how this was built

This project — the scripts as well as this README — was written with the assistance of AI, working against a real photo library rather than a synthetic one. Most of the behaviour documented above exists because a scan of that library turned up something the first attempt got wrong.

It has been tested end to end, but it moves your files. Run the read-only scan first, read plan.csv, and keep the reports\ folder until you are satisfied — that is the only route back.

Licence

MIT