Can AI Analyze a batch of photos but not auto create Category Labels in my Personal Categories?

[tplunkett]

I just upgraded to PS2025. In experimenting with Open AI integration, when I AI analyze a single photo and do not process the results automatically, I like how PS records the category labels in the private AI Labels and does not write them to my image metadata (unless I move the AI Label to my non AI, personal Category labels). What I tried next was batch analyzing 40 pictures and I see that PS states it auto processes when multiple images are selected. I do not like that in this case as it does not create Private AI labels and instead writes them into my personal Category labels. Am I doing something wrong? Seems like I should be able to batch analyze but not have the labels auto added to my personal category labels that I write to my image metadata. Is PS doing this because I have parent categories in my personal labels that match the parent categories that the AI results are classified into? For example I have personal label categories of Events, Objects, Styles, Miscellaneous which match the AI categories that the labels are being assigned to. If so, should I rename my personal label categories to a different name to avoid having these AI labels auto added to my personal categories and then would they remain as Private AI labels?

My category tree: [ external image ]

[tplunkett]

To follow up on this, in the User Guide where it states for AI analyzing a single image:

"Keywords found during processing are unselected by default. For unselected
keywords, the system checks if they match an existing label in that category. If a
match is found, it's assigned to the image; if not, a new label is created in a private
"AI Labels" category. Labels in this private category aren't included in the image's
metadata but are saved in your catalog for later use."

I think right now in batch mode the keywords found during processing are treated as being selected by default and thus they are created in my existing catalog categories and not in my catalog's Private AI label categories.

If someone could validate that this is how it is working by design, I would appreciate it.

If true, I will likely enter an enhancement request to see if in batch AI analysis mode the system could prompt for whether you want the system to treat the AI returned keywords as selected or not selected and then process them into the catalog accordingly.

[andrew.heard]

Hi tplunkett. Pity no one else has replied ;-(
I've just upgraded to PS2025 & setup OpenAI. I haven't tried batch processing, just a single thumbnail so far, but like you, my experience seems "at odds" with the Artificial Intelligence manual PDF. I'd be happy if new categories were assigned in "AI Labels" but instead they are assigned under Events, Objects, Miscellaneous. I then have to check for each newly assigned category & move as appropriate. The white triangle on the top/ left of each label seems inconsistent too - some existing labels have the triangle, so don't.

I was hoping that selecting “Assign Know Labels” instead of “Assign All Labels” would show all labels in the panel, with new ones unassigned, but this isn't occurring.

[tplunkett]

Hi Andrew,

Thanks for your input. I am a little disappointed and surprised that this is not impacting more people or that they do not feel the same sentiments in regards to its functionality in this use case. The way it currently is functioning in this regard and cluttering up my non AI category labels makes it such that I cannot use it. I love the idea of the AI helping me out but in this use case it actually does not help but rather adds extra work and effort such that I will not use this feature.

I was hoping I was not using it properly or there was another way to do this AI Analysis across multiple images at once. I guess the rest of the PS community may not be using it in this manner or feels the way it works is somehow efficient for them in this scenario.

[G8DHE]

I've played with all the various versions and I have to say I have yet to find a real use for the labels it comes up with. Finding them to be so general as to not be of any great use. Most of my labels relate to Locations, Individuals, mechanical parts. Being able to identify all Dusk / Sunset / Outdoors / People / Car etc. isn't so far of great use for my purposes. Possibly it will getter better over time so will keep a watching eye on what level of accuracy and relevance is achieved.

[andrew.heard]

https://www.idimager.com/what-s-new-in- ... preme-2025 contains a worthwhile gem you may have missed, like me: #7538; "Added a bulk AI handling for Catalog Labels to only create them in the AI Labels category" - this means all OpenAI labels are now created under the top level AILabels category - see screen cap. For me, that changes the feature from too-much-work-to-bother-using to a nice-improvement-that-may-save-me-time From a users point of view, some of these 1-liners hide a lot of detail that is difficult to appreciate. I can't complain though, most apps/ programs don't say what has changed at all when a new release occurs; or even worse, just silent auto-update. Hert is exceptional in that regard & should be congratulated.

I've only experimented with two "test" thumbnails so far, but #7538 now gives me a bit more confidence to try some "bulk" operations without cluttering up my existing categories. The labels that OpenAI found were quite interesting, some useful, some not so, but if they are in their own top level category I don't think it matters, to me.

[Mke]

I've played with all the various versions and I have to say I have yet to find a real use for the labels it comes up with. Finding them to be so general as to not be of any great use. Most of my labels relate to Locations, Individuals, mechanical parts. Being able to identify all Dusk / Sunset / Outdoors / People / Car etc. isn't so far of great use for my purposes. Possibly it will getter better over time so will keep a watching eye on what level of accuracy and relevance is achieved.

Conceptually (I can't talk practically yet as I currently only have my laptop, which doesn't have the resources for AI), I suspect that AI needs to go a step beyond recognizing things to be really useful. That step would probably be to 'follow' a user as they catalog their images in 'lean' mode, so that it can then apply that knowledge to catalog other images in a similar way.

[G8DHE]

Indeed, I also find the very flat hierachy they produce not at all useful, all of my labels are very hierarchal so the least that would be useful to me would be "Animals|Dog" etc. otherwise I end up having to drag the label across to the right place in my own structures.

[RobiWan]

I think you have to look at it a little differently at this point - what you enter yourself “properly” hierarchically is great. What the AI finds as an addition - order, no, the AI doesn't know that.

[G8DHE]

Just been testing out some more processing around 1000 images.
One thing rapidly comes to the fore is that the Label panel could do with some additional controls - like ability to turn Off AI labels when necessary (large numbers of them) and maybe to sort the Labels alphabetically even, not sure if Hert can add such controls to the Label panel?