I agree to what you found. The RAW Remove Noise leaves mushy, flat and watercolor-like detailless greens, blues or reds. I prefer what the latest LR does as to noise removal, much clean and detailed images.
When I go through the extra module in Lightroom Classic to send the DNG directly to photo ai and after Iâve made my little adjustments, well I notice that when I save to Lightroom, it loops.
After 5 minutes itâs always like that.
It seems that the function to switch from Lightroom to Photo Ai from the extra module is broken at registration. Or it is to save DNG files that is broken.
luckily, by right-clicking send as TIFF file, it still works fine.
I confirm that as a plugin there is a real problem with DNGs. Because I just checked with a Tiff through the module - extra and everything works as expected.
I havenât tried it in standalone mode yet. But in principle, it should work.
Thereâs still a visible pixel shift when sharpening:
Edit: it only applies to the âStandardâ model. The other sharpening models do not shift.
Itâs normal, itâs with the sharpen strong or with motion blur. Itâs to catch the blur due to the camera movement.
The above is using the normal sharpen, not strong or motion blur.
Strange because I donât have the problem with the standard sharpen
I only have the issue with standard sharpen. The other 3 models donât do this.
Actually thatâll be my workaround, thanks. Iâll just use the strong model on a lower setting. Seems to look better with less blooming too.
Another bug, it adds strongly stretched lines on RAW files of type ORF
Before this kind of bug was on DNG files and not on ORF files.
The good news is that the export still works on ORF files. The bad news is that it also exports with strange lines as if the edges of the image had been stretched. See image on the right
in conclusion, on Raw files, the last version brings a lot of bugs that were not present on the previous versions.
Yes I restarted my machine, I even reinstalled my Mac OS.
Too bad it didnât solve the issue. Hopefully support will step in.
interesting ⊠seems that facial recognition was on and sharpen went around the face mask which is not shifting.
Youâre right, face recovery indeed undoes the shift for the face only. The result is a slight misalignment of head and neck.
Regarding update 1.3.3, unfortunately it is still unusable for my purposes. It is nice to see the randomized/moving blurry patches seem to be gone. However, There are still numerous, stationary blurry patches in my nature photos in both Standard and low rez modes, but now they donât move around and change.
Also, still seeing stringy rendering at times in boulders and rock faces, like they are biological rocks.
As well, I was disappointed that the rendering in the viewport now seems significantly slower. Iâm running an RTX 3090, Xeon Gold with 40 cores and 456 GB memory, and with the previous version, I could pan the view and the 3090 could usually keep up with my panning, rendering within a second or two. Now, it seems like itâs much slower, perhaps several times slower in rendering the view. Just FYI.
Thanks for updating frequently though. I look forward to the end of blurry patches!
I have also noticed that the sharpen pixel shift is sometimes very weird in that some parts of the image are shifted and some arenât. It creates a weird, small distortion. I had not connected it to face recovery though. I will have to check later if the differential in pixel shift only happens when both sharpen and face recovery are used. I first noticed the pixel shift problem last year and it has continued for all releases since then.
Thatâs probably because âsubject onlyâ sharpening is the default. In that case only the (auto)masked subject shifts.
In my testing the shift is not dependent on using face recovery.
Are you saying that there is a pixel shift differential between various parts of the image using only sharpen and not face recovery? Your example seems to show that with face recovery the face does not shift, but the other parts do. You later wrote this:
I noticed yesterday some artifact bugs with the âHigh Fidelityâ model. As you can see on this picture.
It seems to happen if the image has already been processed with 100% denoise from another program such as the new Ai denoise from Lightroom classic.
This disadvantage was not there on the 1.3.2 of Photo AI. Moreover this model was even better in the previous version. So for the images I process, the Standard model is superior in quality.
We can see with the same settings the difference between the 2 models.
Here with High Fidelity
And here with Standard