Skip to main content

Duplicate Detection

If you imported the same video twice, or have near-identical clips from burst mode, you don't want both in your memory video. Duplicate detection catches these before they waste slots.

Perceptual hashing​

Each video gets a perceptual hash: a fingerprint based on what the video looks like, not its file contents. Two videos with different codecs, resolutions, or compression levels but the same visual content will produce similar hashes.

The hamming distance between hashes determines similarity. A distance of 0 means identical. Higher values mean more different.

Hashes are computed on Immich's preview thumbnails, not the full videos. The web UI caches those in Step 1. generate and auto run fetch whatever is missing at the start of the clustering phase, using analysis.download_workers parallel clients (3 by default), so the automated path de-duplicates the same way the UI does. Previews are cached on disk, so a second run over the same clips fetches nothing. If a preview can't be fetched, that clip skips duplicate detection, a warning is logged, and the run continues.

Configuration​

analysis:
  duplicate_hash_threshold: 8  # default

The threshold controls how aggressively duplicates are matched:

  • Lower values (e.g., 4): only catches near-exact duplicates
  • Default (8): good balance, catches re-encodes and slight crops
  • Higher values (e.g., 12): more aggressive, might false-positive on similar-but-different clips

How duplicates are resolved​

When a group of duplicates is found, they're ranked by favorite status first, then resolution, bitrate and duration, and only the top one is kept for selection. A favorited duplicate therefore wins even at a lower resolution — you flagged that copy on purpose. The others are silently dropped: they still exist in your library, they just won't appear in the generated video.