I think the problem is that AI needs a lot of images to train it well, and perhaps they don’t have a big enough library or varied enough user interactions.
1 Like
I am not referring to this particular bug but in general most of the bugs reported should have been found in internal testing. It almost seems as though they are using users to test the software with minimal internal testing. In addition, constantly adding new features gives the impression that it was release prematurely and is not ready for prime-time usage.
windows 11 RTX 3060
Very encouraging
With good cases may be about perfect This one refine with just two clicks on small chunks one add on the bird one subtract on the water
Of course in many other cases will need to have a way to select more adjustable chunks , for smaller chunks to adjust more details and larger chunks to remove faster large parts
In this case not possible to mask correctly the tops of the capitals on the sky and to remove quickly all what is not the church
Of for the future we would like to have subjects detected like bird or architecture
Subject =default
Subject=Paysage
Also probably the mask will need to be refine with colors and luminosity selection , just like Adobe is doing !!!
Interesting to compare with the new automatic Adobe AI mask subject
1 Like
Exemple of adding too small chunks to the default automatic
will be interesting if an architecture subject will looks like this obtain diectly with On1 super select AI
Denoise AI is the left image. Photo Ai is the right image. photo ai noise is set to 1 with maximum detail.
there is a clear loss of detail on the clothing with Photo Ai, Denoise AI however retains more detail. Photo ai’ noise profile is far too aggressive with raw files. Then there is the biggest mystery for me, why are the colours and brightness different in Photo Ai compared to Noise Ai?
2 Likes
I just tried out the masking on quite a “busy” image with difficult transitions. Masking only seemed to work properly when zoomed in on the image. Unable to use the standard Mac keyboard input to move around the image so cleaning up extraneous material was a bit time consuming. By default it had chosen some of the people in the background as well as the main figure, on whom focus had been missed.
I noticed that, when masking is set to subtract, moving the crosshairs down and out to click the Done button is likely to subtract from part of the image if it moved over an area that had been chosen with Add. Screenshot below.
Finally, and I have said this countless times about Sharpen AI, PLEASE PLEASE can we have the ability to save a mask! With this image, after masking and saving the image itself, I realised that I had forgotten to add in the grapes and apples. If I want to do that now, I’d have to start all over again.
1 Like
Well, that’s interesting!
IMO a photo processing program shouldn’t add vignettes unless users specifically ask it to. That is an artistic finishing choice that not everyone wants to add in all circumstances.
In Ps, there is an option - when stitching a bunch of images - to say, “Remove the vignettes from individual captures in the process of creating a pano”. That’s why I wondered if you did an (for lack of a better term) exposure equalizing for a stitched pano & then chopped it up if you could perform the PAI actions w/out enhancing a pre-existing subtle vignette. But if the current rel. of PAI is reintroducing a vignette around newly, randomly created (sized to work) segments of a pano, that does seem challenging for your appln. It would be for my non-drone landscape shots if I tried to work with them like you are…
Take care, Barney. Interesting to hear what you found in your tests!
1 Like
I’m on a PC but found the same. A fairly high priority, besides brush size, is the ability to easily move around the image (Hand func. or Spacebar & drag in Ps) the image to create/refine the mask.
Saving the mask implies (to me) either a layered approach or at least a Clipboard type feature for the duration of the session.
On a try with the new select and brush I tried as suggested earlier and zoomed in close to make adjustments. Tried to press the space bar to move around the image when zoomed in and found it would not work, Have found several basic Windows command formalities are not observed by this software, the area of screen taken is now standard and the previous AI offerings observed it. On one image I tried to use the windows Win Key + Shift + S to copy and area selection of the screen , this first caused a past image to be shown on screen , and then a lockup.
1 Like
Hello, excuse my dumb question, but the subject / mask feature is used only for the sharpen filter, to make sharp only the subject / deselect it for the whole image or it serves other purposes ?
- Default zoom is now “Zoom to fit” on small images
don’t know what “small images” means to PhotoAi, but would be good to set up this for all the pictures in preference or turn it off if we want.
I noticed that if i move the zoom setting of the picture while it’s analysing, and turn on enhance / scale, sometimes, the picture is zoomed at full and 100% become the default. if i want to zoom out it’s impossible, the “fit” become something unusable like 247% or something like that.
only way to reset this is to click on scale x1 and enter again the desired height (i have 2160px as default height in the preference).
The latest updates are great. i agree that the subject brush tool needs lots of improvement. I have got spoiled with Lightrooms tools that can be adjusted by size and edge adjustment. keep it up
1 Like
The problem is Global vs Local adjustments. No doubt that Photo AI does a great job of enhancing detail in each frame, but the resulting vignette at the edge of each frame is a killer. I guess that, if I masked out the areas NOT to be touched by AI, that would work, but that’s a major job.
As far as stitching is concerned, PS does a very simple stitch, whereas PTGUI permits absolute control.
1 Like
I got the Remove Brushed on Mask (using the right click button) to work. I am pretty excited!
I found that if I did what others above suggested and zoomed in a lot the edge detection definitely stayed more within the subject I wanted it to.
I did that work using the PAI 1.0.6 Ps 2022 (not updating for a couple wks till others kick the tires of 2023 …) Plugin.
The plugin is launching to full-screen (yay!). But, the processed pic returned to Ps layer still shows a pixel hop when switched on/off.
Hello,
I found a possible degradation on v1.0.6, comparing to v1.0.5.
If I choose multiple photos more than 12 on lightroom, then Photo AI just read 12.
The attached pic shows I choose 26 photos on lightroom, but Photo AI just show 12 photos.
It worked fine at v1.0.5.
I usually process multiple photos around 50, so it helps me a lot if you fix this.
My environment.
OS: Windows11
CPU: Core i7 12700KF
Memory: 64GB
GPU:Nvidia GeForce RTX 3060
Lightroom v12.0.
Thanks in advance!
After a bit of experimenting I can move around the image when masking by holding down the spacebar and dragging - but I can only do so after moving the box in the navigator panel at the top right.
1 Like
I have been taking another look at some images from a trip a few years ago that I underexposed. Deliberately so, because of the hot desert sun, so I didn’t want to bleach out the sky. I know the consensus a few iterations back was that tone adjustments were not needed in PAI. But I’d like to request an exposure compensation slider. For no other reason really than it makes adjusting an image manually difficult when much of it is too dark.
Anyway, for what it’s worth, PAI didn’t do a very good job and introduced some nasty artefacts.
I also tried DxO Deep Prime and On1’s built-in denoising and sharpening. But by far the most acceptable result, yet again, was with Topaz Sharpen AI.
-
Edited in On1 Photo Raw using both NoNoise AI and TackSharp AI:
-
TPAI on Autopilot (I suspect sharpness could be improved but note the artefacts
-
DxO DeepPrime and then sharpened in On1 PR
4.Topaz Sharpen AI, in auto. Not perfect either, but then my lens was not top of the range.
On my Apple trackpad, I can use gestures to move around an image and zoom in and out.
Update to my tests on underexposed images. TPAI did much better when I switched sharpening from Standard to Lens Blur. It always seem to select Standard as standard!
1 Like
