The problem I have is that there's no way to embed an "uncertainty" into EXIF metadata in a standard way. I just want some photos to be "Summer 1987" or "February 1976" or even "1981-1983"... but I _have_ to invent some complete timestamp down to the second and then just rely upon captions or comments.
But I generally know the dates far better than any AI could guess (based on ages of the individuals I know).
You supply the ages with the software and AI uses that just like you would.
If it doesn't know the month, it puts it in June. If it doesn't know the day, it puts it on the 15th. But yeah, old photos with EXIF is more to be used by putting them in an order of photo1 comes after photo2, etc. Not necessarily photo1 is exactly 1982, etc. Viewing and enjoying old family photos is more about the order, and less about the exact year (at least for my family and friends).
OCRing handwritten dates and printed timestamps seems useful, but how is “AI” supposed to know when a photo is taken just by looking at it?
At least the scan date is a real piece of data that could be useful. This seems like it would cause more harm than good by polluting the data with nonsense, unless the added dates are clearly labeled as estimates
It does well with visual queues in the photo if there is no handwriting for dates or timestamps. It's definitely not perfect but it will get it so your photos feel chronological in the sense of "photo A is older than photo B".
The way to increase the accuracy is to add a birth year to face tagging so each photo has an anchor that the AI uses -- "This photo has joe, joe is born in 1950" and then it will look at how old joe is and if joe looks like 10 years old, it knows the photo is around 1960.
Seems like a great idea for an Immich plugin. I'm really not convinced I or anyone I know would _pay_ for AI to guess the date a photo was taken though.
I do integrate with Immich. It will find photos with missing EXIF data automatically for you.
It will also only write back metadata to update the photos in place, so you don't have to do the upload-download-reupload dance.
> Google Cloud (Gemini Enterprise): AI date extraction. Photos are sent to Gemini Enterprise solely for date extraction and are not used to train Google’s models.
Microsoft PhotoDNA: child-safety hash scanning (see “Child safety scanning”).
Scanned photos all show up on the day you scan them. Timeline Scan estimates when each photo was actually taken and writes it into the file’s EXIF, so your archive sorts chronologically in Immich, Google Photos, or Apple Photos (or by folders if you’re one of those people). I increased the trial to 50 photos; try it out and let me know what you think!
It does both. Each image and then neighbors. Usually you scan photos that are printed in related-date order. So it does look at neighboring photos to date.
You can tag faces and add a birthdate; so sort of. Like it'll recognize a birthday party photo and then if it's tagged with that person in the picture, it will date it on their birthday.
The problem I have is that there's no way to embed an "uncertainty" into EXIF metadata in a standard way. I just want some photos to be "Summer 1987" or "February 1976" or even "1981-1983"... but I _have_ to invent some complete timestamp down to the second and then just rely upon captions or comments.
But I generally know the dates far better than any AI could guess (based on ages of the individuals I know).
You supply the ages with the software and AI uses that just like you would.
If it doesn't know the month, it puts it in June. If it doesn't know the day, it puts it on the 15th. But yeah, old photos with EXIF is more to be used by putting them in an order of photo1 comes after photo2, etc. Not necessarily photo1 is exactly 1982, etc. Viewing and enjoying old family photos is more about the order, and less about the exact year (at least for my family and friends).
OCRing handwritten dates and printed timestamps seems useful, but how is “AI” supposed to know when a photo is taken just by looking at it?
At least the scan date is a real piece of data that could be useful. This seems like it would cause more harm than good by polluting the data with nonsense, unless the added dates are clearly labeled as estimates
It does well with visual queues in the photo if there is no handwriting for dates or timestamps. It's definitely not perfect but it will get it so your photos feel chronological in the sense of "photo A is older than photo B".
The way to increase the accuracy is to add a birth year to face tagging so each photo has an anchor that the AI uses -- "This photo has joe, joe is born in 1950" and then it will look at how old joe is and if joe looks like 10 years old, it knows the photo is around 1960.
Yeah, seems like snake oil with a dash of malarkey.
Seems like a great idea for an Immich plugin. I'm really not convinced I or anyone I know would _pay_ for AI to guess the date a photo was taken though.
I do integrate with Immich. It will find photos with missing EXIF data automatically for you. It will also only write back metadata to update the photos in place, so you don't have to do the upload-download-reupload dance.
> Google Cloud (Gemini Enterprise): AI date extraction. Photos are sent to Gemini Enterprise solely for date extraction and are not used to train Google’s models. Microsoft PhotoDNA: child-safety hash scanning (see “Child safety scanning”).
Technical background.
I work at a large social media company that handles photos. :salute:
Scanned photos all show up on the day you scan them. Timeline Scan estimates when each photo was actually taken and writes it into the file’s EXIF, so your archive sorts chronologically in Immich, Google Photos, or Apple Photos (or by folders if you’re one of those people). I increased the trial to 50 photos; try it out and let me know what you think!
Based on what? Each image individually? Or does it try to cross-reference images within the collection and with events in your life, etc?
It does both. Each image and then neighbors. Usually you scan photos that are printed in related-date order. So it does look at neighboring photos to date.
You can tag faces and add a birthdate; so sort of. Like it'll recognize a birthday party photo and then if it's tagged with that person in the picture, it will date it on their birthday.
I think this method could do well with handwritten journal entries and artists’ sketchbooks, that add another dimension of information.
haircuts
doesn't handle 80's themed parties very well.
vibes