Fabio Manganiello on Nostr: I wish I discovered #Immich earlier. For the past couple of years I've tried to get ...
I wish I discovered #Immich earlier.
For the past couple of years I've tried to get #Nextcloud to manage my photo collections (~25k items accumulated over many years).
I have tried both the Memories and Photos apps, but none could even remotely do a decent job.
My Nextcloud server with its 8 GB of RAM would freeze, swap and become unresponsive for minutes as soon as I started to scroll a collection. And any operation more complex than basic transcoding and thumbnail generation (face detection, geo-tagging, duplicate detection...) wasn't even an option.
Nextcloud photos apps don't seem to implement proper caching, nor pre-processing, and they also try to transcode and convert media files on the file while the user scrolls, apparently without limits to the number of processes.
Immich instead seems to provide exactly what I was looking for. < 1 GB of RAM usage even while transcoding, embedded lightweight models for facial recognition, configurable parallelism for the processing jobs, and lots of pre-processing to make the experience as smooth as possible when scrolling and searching.
For the past couple of years I've tried to get #Nextcloud to manage my photo collections (~25k items accumulated over many years).
I have tried both the Memories and Photos apps, but none could even remotely do a decent job.
My Nextcloud server with its 8 GB of RAM would freeze, swap and become unresponsive for minutes as soon as I started to scroll a collection. And any operation more complex than basic transcoding and thumbnail generation (face detection, geo-tagging, duplicate detection...) wasn't even an option.
Nextcloud photos apps don't seem to implement proper caching, nor pre-processing, and they also try to transcode and convert media files on the file while the user scrolls, apparently without limits to the number of processes.
Immich instead seems to provide exactly what I was looking for. < 1 GB of RAM usage even while transcoding, embedded lightweight models for facial recognition, configurable parallelism for the processing jobs, and lots of pre-processing to make the experience as smooth as possible when scrolling and searching.