I am using both PC and MAC and the problem is the same on both. When I am sending an image from Lightroom to Photo AI using Plug-In Extra I very often find that detailed parts of the image is kind of smeared out or blurry, this should be easily seen in the three screenshots. Photo AI is set to only reduce noise on autopilot but even if all settings are reduce to minimum, nothing changes. I have tried to use manual sharpening but it has no effect. Using Edit In from Lightroom gives a better result and using Photoshop gives a perfect result.
There are no adjustments made in Lightroom other than setting sharpening and noise reduction to 0 (zero).
The attached screenshots will probably show my problem.
I too have found this to happens on smaller delicate structures especially with red colors softening greatly as soon as I enable noise reduction (even at the lowest levels of NR) it just had to enabled in PhotoAI to see the smear effect. I cant wait to see how this resolves… I submitted my image to the team but I didn’t spend the time to do all the screen shots like you. fingers crossed.
If you want to retain the most detail when upscaling the image, I recommend using the High Fidelity model for Enhance Resolution.
It looks like the starting image has texture due to low quality. When upscaling with Topaz Photo AI, some of that texture is removed with Enhance Resolution to smooth the image and increase the quality.
Regardless of the model (Standard, High Fidelity, Graphics, Low Resolution), the same issue happens. Blurry or sharp spots appear seemingly randomly. Images were reduced in quality when I uploaded them, thus making it very hard to see the difference. The original images are from Midjourney in PNG format and have no flaws. The texture of the face is the same all over. These issues did not happen before with the natural model that is now removed. Seems like all existing models have an issue with focal depth, thus suddenly sharpening or blurring the image. Maybe the old natural model handled this differently? Would it be possible to implement the natural model again?