[Question] How Does Immich Select Video Thumbnails? (Deduplication Issue with Google Photos Imports)
Hey all,
I've been migrating my library from Google Photos using Google Takeout and uploading everything to existing Immich via the CLI.
In going through my Immich library, I'm trying to clean up duplicated videos—particularly where both the original and the Google Photos–exported (compressed) version exist. Immich's duplicate detection is doing a great job with photos, but with videos, it’s much less reliable.
Here's my observation:
Two videos with identical content and length (original and Google Photos–exported), but different file sizes/codecs, don’t get the same thumbnail.
Comparing the thumbnails, it appears that the larger file's thumbnail is taken a bit earlier (by a few hundred milliseconds) than with the compressed Google Photos file. The result: thumbnails look different, even though the videos themselves are visually identical when played.
Because of these thumbnail differences, Immich's deduplication isn't flagging them as duplicates.
My questions:
How, exactly, does Immich decide which frame to use for video thumbnails? Is it always at the exact first frame, first keyframe, or does it attempt to find a "representative" (non-black) frame further in?
Is this frame selection affected by the file’s encoding/compression/keyframe map, or by timing offsets introduced by transcoding when exporting from Google Photos?
Has anyone figured out a reliable way to force Immich to generate matching thumbnails for such duplicate videos—or is there a way to base deduplication on something more robust than the thumbnail?
Any tips (regenerating thumbnails, settings tweaks, scripts, etc.) for deduplicating videos imported from Takeout, where the original and exported files are visually identical?
Thanks for any advice or technical detail you can share!
I've been migrating my library from Google Photos using Google Takeout and uploading everything to existing Immich via the CLI.
In going through my Immich library, I'm trying to clean up duplicated videos—particularly where both the original and the Google Photos–exported (compressed) version exist. Immich's duplicate detection is doing a great job with photos, but with videos, it’s much less reliable.
Here's my observation:
Two videos with identical content and length (original and Google Photos–exported), but different file sizes/codecs, don’t get the same thumbnail.
Comparing the thumbnails, it appears that the larger file's thumbnail is taken a bit earlier (by a few hundred milliseconds) than with the compressed Google Photos file. The result: thumbnails look different, even though the videos themselves are visually identical when played.
Because of these thumbnail differences, Immich's deduplication isn't flagging them as duplicates.
My questions:
How, exactly, does Immich decide which frame to use for video thumbnails? Is it always at the exact first frame, first keyframe, or does it attempt to find a "representative" (non-black) frame further in?
Is this frame selection affected by the file’s encoding/compression/keyframe map, or by timing offsets introduced by transcoding when exporting from Google Photos?
Has anyone figured out a reliable way to force Immich to generate matching thumbnails for such duplicate videos—or is there a way to base deduplication on something more robust than the thumbnail?
Any tips (regenerating thumbnails, settings tweaks, scripts, etc.) for deduplicating videos imported from Takeout, where the original and exported files are visually identical?
Thanks for any advice or technical detail you can share!
