I'm using this one. $12 lifetime license.

Media Insight – Digital Forensic Image & Video Analysis Tool
https://www.videomap.it/mediainsight/#features


Like the above, /r also turned up these two;

Features of JPhotoTagger
https://elmar-baumann.de/JPhotoTagger/features.html

GitHub - DIVISIO-AI/stag: An AI based automatic image tagger
https://github.com/DIVISIO-AI/stag#readme

I'm running win64 Stag. It has a current log that shows what kws it has posted to the XMP sidecar file.  I loaded a folder with a copy of my photo project -2850 jpgs. Many were scanned or taken with early digital cams, so it's having to create the sidecar. Processing rate is roughly 40-50/min.

Next, I'll try to write the Stag tags into the image itself and dump the sidecar.

For those reading along at home, here's a sample of the generated tags per image

Tags:  'belly', 'bathroom mirror', 'catch', 'smartphone', 'check', 'doorway', 'person', 'man', 'phone', 'photo', 'picture', 'pink', 'selfie', 'shirt', 'stand', 'stomach', 'take', 'woman'

Tags:  'arm', 'swimwear', 'boy', 'child', 'hand', 'pool', 'shirtless', 'stretch', 'swimming pool'

Tags:  'armchair', 'baby', 'chair', 'child', 'table', 'dinning table', 'feeding chair', 'fireplace', 'floor', 'lap', 'laptop', 'living room', 'man', 'picture frame', 'play', 'room', 'sit', 'stool', 'toddler', 'toy', 'woman', 'wood floor', 'wood wall'

Tags:  'blue', 'boot', 'cowboy boot', 'dress', 'girl', 'hair', 'sandal', 'shirt', 'short', 'stand', 'wear', 'woman'

Tags:  'catch', 'child', 'daughter', 'electronic', 'father', 'girl', 'ipad', 'lap', 'man', 'tablet', 'play', 'tablet computer'

Tags:  'swimwear', 'bend', 'boardwalk', 'check', 'child', 'dock', 'girl', 'help', 'learn', 'man', 'sea', 'stand', 'teach', 'water', 'woman'

Tags:  'swimwear', 'bend', 'boardwalk', 'camouflage', 'child', 'dock', 'girl', 'help', 'learn', 'life jacket', 'man', 'sea', 'teach', 'water', 'woman'

Tags:  'swimwear', 'beach chair', 'chair', 'child', 'deck', 'girl', 'house exterior', 'porch', 'rail', 'rocking chair', 'sit', 'white', 'woman'

Tags:  'brunette', 'couch', 'dress', 'girl', 'laugh', 'pose', 'selfie', 'sit', 'smile', 'woman'
 

It had to create an xmp file for every image. It averaged 1250 files with tags per hour.

In my batch it created a total of 34k tags, an average of 12 per photo.

Many of the tags were used repeatedly, there were 1600 unique tags.

exiftool -tagsFromFile %d%f.xmp "-IPTC:Caption-Abstract<XMP-dc:Subject" *.jpg

copies to images

+
Works. Done.

What do I lose if I skip all this and just use Qsirch?

It's interesting - I feel like that summarizes the big question of our time. Do I need to be running a tool to tag this and that and then add exif tags, or do I just offload this to an blackbox AI and hope for the best?

All these tools look interesting, but the realistic person inside of me says that I'm more likely to actually use Qsirch if it's even halfway decent

>Qsirch

Have you installed the free version?  I have not.

My 1st question would be can it write to the jpg fields or does it just keep a separate db?

2nd: How does it name the individual's face tag with the familiar, searchable name the user wants?

As mentioned, I ran STAG and it did indeed add some usable, mostly correct tags. BUT some were incorrect (boy for a girl, etc) and others were pretty damn obscure and likely never to be used.  I've written scripts to clean them up.

On family photos I would not just turn AI loose hoping for decent results.  But on 7800 travel photos? Yeah, I might --but only if it wouldn't overwrite my manual tagging.  I'd already decided to run STAG on my travel directory.  In this case, some ridiculous tag might just be how I remember one cathedral vs another (out of seemingly hundreds, I got tired of cathedrals & castles).