Thanks @Red, you prompted me to do more experimenting.
Yea, I can’t really get 5.5.2 looking good. But in 5.4.5, for some reason, GPU/OpenVino turned off looks better than GPU or OpenVino (which look the same when A/B’d). notice the hair and the dress especially. It even looks different in the preview than exported. I’m frustrated by the lack of consistency.
Here is the original file and my settings if you guys want to test this yourselves.
5.5.2 is really bad. It destroys my Face. Check the eyes. Also eyebrow is missing detail. Plus rest details are gone. U suck Devs. Why every update and soft u update is from bad to worse.
Here i got 4 pictures to compare the versions
Standard 5.4.5
Standard 5.5.2
Lines 5.4.5
Lines 5.5.2
Well, I think the preview itself is not useable anymore. Maybe avoid to look at the preview itself. Just process the image and compare after. 
Did you ever try the “lowRes Model” for low res images?
I’m still on 5.1.7, and ran your original using my GPU. It came out looking sharper and cleaner than any of the other examples we’ve been looking at. The only negative I noticed was a bit more color artifacting.
Speculation… it appears the new versions have the sharpness reduced. I’m left wondering if that’s because it’s felt by Topaz that this looks more natural… or, because if people want more sharpness, they should use SharpenAI (?!?)
Anyways, there should be an easy option for people who want results that are comparable to 5.4.5 or earlier.
More sharpness = more artifacts. Pictures should not be too sharp, or it will look so unnatural.
A subjective opinion: Gigapixel AI and Video Enhance AI results are looking more artificial these days. Sharper, cleaner, and yet less real and more artificial. It seems artificial intelligence is showing its true colors. 
Also in my completely subjective opinion, they definitely do destroy fine detail to some degree and introduce weird edge distortion due to AI over-sharpening unfortunately.
Generally, they provide satisfactory results only if the source is good enough i.e. well lit shots with little to no fine details, otherwise I don’t bother with them really…
I honestly think that it’s just due to the new engine/fast models and as a result, the reduction in quality comes with the territory. I’m fairly certain you would get similar results with 5.4.5 as well if you disable the “use maximum quality AI models” option.
Yes, I agree with you, Francois.prioux.
“Also in my completely subjective opinion, they definitely do destroy fine detail to some degree and introduce weird edge distortion due to AI over-sharpening unfortunately”
This is exactly what I think, most of the time. It’s like the algorithm decides to overdone some areas for some really strange reason, to the point that everything becomes wet and/or plasticky, so to speak. Maybe this algorithm has some “redundancies” that need to be fixed? I don’t know, but this is really strange indeed.
I ran your original picture through Topaz Denoise first then opened it in Gigapixel 5.5.2 . I ran it with CPU (I don’t see any Open VINO so I don’t know if it is used) and then my GPU Radeon RX5600 XT. I think both look good but the GPU shows a bit more detail on the skin.
CPU 2X Low Resolution:
GPU 2X Low Resolution:
Sorry mate, completely forgot to answer your question. 
These are the most optimal settings for both version:
5.4.5:
5.5.2 (With all models deleted from tgrc excpect for tensorflow (192x192-tf.tz only)):
Settings used in the comparison picture: Standard Model, Auto Settings, Reduce Color Bleed On.
Using legacy tensorflow (openvino off) and your settings:
5.4.5:
5.5.2:
I tried every single combination possible but the app settings I provided above are the ones that yield the best results.
hm it might be a rtx thing then. thanks for posting these.
yup, as i said, im sad that theres no consistency here. thanks for posting
i tried all combinations in 5.5.2 and it still ended up blurry, but you can try red’s legacy settings because they seem to turn out pretty well
Okay, foolishly, I installed 5.5.2 forgetting that I’m actually taking part in beta testing here…
I noticed that downloading modelling was stuck at 1%, so I restarted and disabled that.
I then chose ALL GPU (previously it was on CPU) to make use of my 2 x RTX 3090 and my machine bluescreened when trying to preview…
I then restarted and retried and chose the 2nd GPU (RTX 3090) and preview worked, but when I went to process a 5000x2500 px PNG to become twice it’s size, it BLACK SCREENED (I have NEVER seen that before).
So… Topaz… tell me… what am I paying for here? Please remind me…
(system is i9-7980xe, 128Gb Ram, 1Tb M.2 OS and 2 x RTX 3090, latest drivers and BIOS).
What psu do you have?
