P: Generative Remove Feedback (Lr Classic & Lr Eco)

Gen Remove Feedback, LrC  14.2

MacBook Pro M3 Max, 64GB 1TB SSD, Sequoia 15.3.2. Experience: LrC 18 years / PS 25 years.

I'm finding all the Gen AI tools immensely frustrating in both LrC and PS. Even worse is the loss of functionality we had previosuly, so I can't even just ignore them. I rarely have issues learning new software but this has ruined my workflow and caused massive loss of productivity. The remove/clone/heal tools in LrC were never brilliant, but they did have their uses for isolated areas away from high contrast elements. Now even that basic functionality has vanished and I have to go PS for 99% of heal / remove clone actions.

File 1 (attached): Task: Remove door knob from foreground. Initially tried Generative Remove but it added so many other artefacts I gave up. Returned to file later to try other tools (Remove, Heal). Results:
File 2 (attached): Fill > Remove. Tried 3 x refresh, difference barely visible, so not included here.

File 3 (attached): Fill > Heal (this would have worked OK previously).

File 4: (immediately below) Refresh x 1 (Seriously? Why sample the curtain? It bears no reseenblance to either doorknob or cupboard and is nowhere near either in image).

File 5: Refresh x 2

File 6:  Refresh x 3

File 7: Fill > returned to Generative Remove. This time it worked. Wish I'd kept all the versions with artefacts it added previously, which included adding copies of the cupboard handles in background, completely new cupboard handles it made up, more versions involving the curtains and removal of white line seperating wardrobe doors.

It took a total of ±12-15 attempts before achieving the final result. I don't have time for this.

I wish Adobe had either left the whole AI thing alone or made sure it worked reliably before integrating it it into every element of the software, and left the old tools alone so users could fall back on them when the AI fails - which is provign to be most of the time so far. It works well for very specific tasks which look great when demonstrated at shows / online, but it's failed me utterly in my everyday work).

First of all, I can't say how amazed that I am at how well this works. I use it often and it's amazing. 

Two things I would like to see. 

1. A way to remove window glare and reflections while still preserving the underlying image. I sometimes shoot from aircraft in varrying different lighting conditions. Recently we flew across the rockys and I took a bunch of shots out my window. I was trying to remove the larger reflections of inside the cockpit without changing the mountains outside, but as the AI uses surrounding areas and features to generate the fill image (at least I think this is what it's doing) I lose the integrity of the original scenery. As much as I usually like the result, I prefer to maintain the original integrity of the photo because I feel that I'm creating a historical record and want it to be representative of the scene I'm seeing, instead of artificiallly generated details.

2. I'd like to see a historical save of the generations of the AI.. There are times I'm hesitant to click the Generate button after looking at the 3 selections because maybe I've gotten the best one within those three, but don't know it yet.

Thanks! Keep up the great work with this. 

I have been using Generative AI to remove people from nighttime and interior displays (Smithstonian Institute); however, a darkish smudge remains after the event. It works well in daylight situations, such as the Arizona desert.

By creating a dust map (F22 of evenly lit surface out of focus), then use visualise spots, can you then generate a slection that can be saved as "Dust Map". Then instead of removing each and every spot manually, the selection can the be filled using generative AI to remove all the dust spots. From the attached image, you can see what a pain it is to remove dust spots manually.