Leaving Google Photos

Google Photos can be left, but what it hands you is a mess. This is the whole path from export to a library you can actually search, written for someone who has not used a terminal tool for this before.

There is no shortcut through an API: since March 2025 Google no longer lets other applications read your library, so Takeout is the only route out. That is fine. Takeout gives you the actual files, and files are what videre works on.

1. Ask Google for your photos

Go to takeout.google.com, deselect everything, then select Google Photos only.

Choose .zip, and set the file size to the largest offered. Fewer, bigger archives are much less tedious than dozens of small ones.

Google emails you when it is ready. For a large library this takes hours, and sometimes more than a day.

2. Extract it

Download every archive and extract them all into one folder. If Google split your library across several files, they are meant to be merged back together.

mkdir ~/Takeout
cd ~/Downloads
for z in takeout-*.zip; do unzip -q "$z" -d ~/Takeout; done

You should end up with ~/Takeout/Google Photos/ containing folders like Photos from 2019 and any albums you made.

3. What Takeout got wrong

Look at the extracted folder and you will see two problems.

Every photo has today’s date. Your capture dates are not in the files. They are in the .json files sitting beside them, and nothing else reads those.

Photos appear several times. A photo in three albums is exported three times, so a 40 GB library can extract to considerably more.

Edited photos appear twice. Every photo you edited in Google Photos comes as the original, IMG_1.jpg, and Google’s render of the edit, IMG_1-edited.jpg, side by side.

videre fixes both.

4. Restore the real dates

videre import ~/Takeout --dry-run

This changes nothing. It reports what it found:

Google Takeout at ~/Takeout
  12,431 media file(s) in 84 folder(s)
  11,902 matched a sidecar (95.7%)
     529 unmatched, left untouched
   8,113 would have their date corrected

The percentage is the number to watch. Above about 95% is normal. Far below that suggests something unusual about the export, and is worth asking about before continuing.

When it looks right, run it for real:

videre import ~/Takeout

It asks for confirmation before changing anything, because this does modify your files: it sets each file’s date from its sidecar. Nothing else about the file changes.

Why some files are never matched

Screenshots, some videos and a few edited copies arrive with no sidecar at all. Those keep whatever date they have. It is not an error, and the count is reported so you can see how many.

5. Build the library

videre --library ~/Takeout scan

This records what you have in a database at .videre/hashes.db inside the library. It reads every file, so on a large library it takes a while. Your photos are not modified.

6. Remove the album duplicates

videre dedupe --html                  # look at what would go
videre dedupe --remove                # move the copies to the trash, once you agree
videre prune                          # tidy the database afterwards

videre dedupe --html opens a page in your browser showing every duplicate group, with KEEP and REMOVE badges. Look before you delete.

This is where the album duplication disappears. Those copies are byte-identical, so removing them loses nothing at all.

Caution

videre dedupe --remove deletes immediately once you confirm. Run videre dedupe --html first to review, or preview with --remove --dry-run. Prefer --remove over a | xargs trash pipe, which splits Takeout paths on their spaces. See cautions.

Then the edited copies

The -edited renders are not byte-identical to their originals, so the step above leaves them. videre import counts them and says so; remove them with:

videre dedupe --edited --html         # each original beside its edit
videre dedupe --edited --remove       # keep the originals, trash the edits

The original is kept because it is the file the camera wrote; the edit is Google’s re-encoded copy. Faces you named on an edit are not moved to its original. See dedupe --edited.

7. Make it searchable

Optional, and the reason to bother with any of this:

videre embed                   # one-time, downloads about 1.5 GB
videre search "sunset over water"

videre faces                   # one-time, downloads about 180 MB
videre gallery          # name the people
videre search --person "Alice"

Both steps take hours on a large library and can be interrupted with Ctrl-C at any point; rerunning continues where it stopped.

Where you end up

A folder you own, on a disk you control, with no account and nothing syncing. The photos are ordinary files: if you stop using videre tomorrow, they are exactly as they are now.

videre stats

Afterwards

Keep the Takeout archives until you have checked the result. Once you are satisfied, they are just a duplicate copy of what you already have.

If you add photos later, the same three commands bring them in:

videre --library ~/Takeout scan
videre embed
videre faces

See workflows for what needs what.