P: Automatic image tagging

In 2026 it is entirely possible for Adobe to implement a local AI agent that looks at our photos and existing key words to then suggest keywords for the photo we’re viewing. 

Ideally it would show a list in alphabetical order of the suggested keywords and allow user to globally “Select all” and “Select None” while individually toggling for individual keywords.

I’d expect AI to emphasize the prior photo’s keywords and make an educated guess on the photos for the current photo. Generally the keyword lists would align with small changes (maybe) if elements changed slightly from photo to photo. For example, if you use a keyword “Ball” and the baseball (football, etc.) goes out of frame, AI should learn to recognize a ball and add or drop “Ball” from the suggestion list. 

AI should, of course, remember keywords used on prior photos and point out elements that are not in photos close (by folder or date) to the current one. A passing butterfly should be captured in keywords because it makes the photo unique within an photo series.

LR Desktop / Mobile / Web offers AI-based keyword search as its "Sensei" search.  But Adobe has never implemented it for LR Classic.

You might consider the Any Vision plugin, which uses Google Cloud Vision, or Excire Search. Any Vision uses Google Gemini AI, letting you write "prompts" that can generate keywords, descriptions, categorizations, and identify GPS locations.

I hope Lightroom can use AI to detect objects/themes/styles in photos and automatically add keywords.

A nice feature would be adding an object search feature in LrC.

How many times have we taken pics and there was no sign of the object name. This could apply to almost any object ie flower, car, vegetable . . .

Current process is exporting a JPG file and loading into Google in Lens search. 

Im LR Entwicklermodus kann das aktuelle Bild automatisch Verbessert werden.
kann diese Funktion auch auf alle markierten Bilder angewendet werden?

"All this presupposes that I have a GPU powerful enough to make local autotagging perform well."

Most common desktops and laptops are more than powerful enough to execute image classification models. In another context, I timed a popular open-source image-classification model:

https://community.adobe.com/t5/lightroom-classic-ideas/p-allow-users-to-disable-the-gpu-for-ai-masking/idc-p/13357752#M17183 

On a 2019 Intel Mac, it classified photos at about 111 photos/sec using its (now older) GPU, and at about 26 photos/sec not using the GPU.  That's more than fast enough and allows plenty of headroom to incorporate larger models.  Excire uses similar technology.

Training new models is much more compute-intensive in general. I don't have enough experience with that to comment on your desire to train on just your own photos. 

After a lot of thought I decided that I did not need Excire or any other tagging software that generatged generic tags such as "boat" or "mountain."  I already do a lot of manual tagging.  I tage faces of course, but I also tag subjects that are of interest to me and that sort of tagging is something that Excire can't do now, and probably won't do for a long time. 

As a Lightroom Classic user (only!) right now, what I need is AI that I can train on MY images only, and use that to tag new photos. And I want it to be Classic only, so I don't depend on an internet connection.

Also if Lightroom Classic did implement auto-tagging, I would have two concerns:

1. Do I even want to include more of the resulting tags?  Tag clutter is a concern for me.

2. How would the auto-tagging align with the keywords I have already created?  Would I need to revise those existing keywords to align with the auto-tagged equivalents.

All this presupposes that I have a GPU powerful enough to make local autotagging perform well.  And GPU boards now are $$$

Will it ever arrive?

I'm wondering whether it would be possible to extend LR's facial recognition to a general subject recognition for things such as "plants", "birds", "flowers" etc.

It would allow one to find things more quickly and greatly reduce the time spent on keywording. I 'm not after 100% accuracy but would happily to settle for 80-90%.

The Apple Photos app does an impressive job at this already.