It’s been weeks now since I’ve inquired about this problem and I’ve watched a lot of updates come and go with no mention of fixing the blurry patches issue. Can we please get an update on how progress is coming on this concern? Are the developers working on it making any headway at all?
Thanks.
I see now. No answer gives me my answer. You are not working on it, or maybe there is no solution.
What bothers me, is so few forum members are chiming in about it. There’s always blurry patches in every photo one blows up, at least of non face type images. The blurry areas are often extremely blurry right next to a perfectly sharp section. Guess I’m the only person out there who wants sharp images all the way across. 
What a stupid childish rant as you raised your post 2 days previously on a weekend. If you search the forum, in May they had said they were working on retraining the models and it would take some time.
I initially wrote back a defensive post, much like a petulant, stupid child might do. But, then I thought to myself, what would Jesus do in this situation? So, like the Grinch who’s heart grew three sizes that day, I’ve come to acknowledge that you were spot on and I was childish and cruel in my writings. I just feel terrible and saddened by my lowly behavior. 
So, thank you dear brother for showing me the error in my ways. Henceforth, I will never behave like an insolent brat from hell again, and rather will turn to the light, and to AiDon, my personal beta testing hero who brought me out of the darkness, showing me my sins like a mirror to my soul. I feel born again! 
God bless AiDon
Love and peace
I appreciate your response as Moderating any forums is a thankless job and sometimes we just vent … I also know that Topaz would not say that they are working on it if they weren’t and, because of experience elsewhere I know training AI models is a very intensive task and needs to be done over and over again for validation.
I am working with the researchers on an update for this. It’s unfortunately not about a fix as we are still exploring solutions.
However, it should at least clear up what the issue is and hopefully shed light on the complexity of it so you and other users are in the loop about what is happening.
It may also help to show you what to look out for in your images to avoid the problem.
That will come this week.
Lingyu,
Really, that’s all I ask is to have communication from time to time on this specific issue. Heck, I’ll even pay money for direct email updates about it.
As I mentioned to AiDon in a Private message, many users may just utilize Photo AI at a hobby/fun level. For me, it’s serious business and holding back a major project, and there’s nothing else like your program on the market, at least quality wise to turn to. So, I’m just sitting here frozen on my project. Thus, updates on progress would be wonderful. Otherwise, I’m left wondering if and when it might ever work right.
If you don’t recall, I’m blowing up tens of thousands of images for a massive gigapixel, so I don’t really have the choice of images to avoid, as the parts make up the whole. I have learned through trial and error how to get the best results from the models, blurry patches aside.
For instance, not using images that are too sharp. I need a bit of blur to maximize the ai results. Same with not too blurry. But, the blurry patch issue, I’ve never found a way around that, save using detail or sharpening enhancement turned up, and those options once dialed up, ruin the images completely, for my purposes. Nor are the images balanced when I try that sort of thing. The problem is in the models as you know. Simply using the standard setting often gives me wonderful results, but again, mixed in with blurry patches and furry rocks.
I should be able to improve communications on these issues now. It took a while to get caught up on the discussion happening in forums.
The difficulty with the blurry patches is that the AI models determine the foreground and background based on context in the block. If two blocks that are next to each other have differing context (one is treated like background and one is treated like foreground) then the tiles are processed/sharpened differently and do not look continuous.
This is a known issue with all generative AI models, not just ours.
Thanks for the update and that’s generally what I felt was going on. I hope there’s a remedy at some point. Let me know if my offering sent to you via PM will help any?
Thanks,
Deven
I asked the researchers what information they would need for this type of work. I will reach out to you once I hear back from them.
For this image I would recommend turning off Face Recovery. This AI filter will try to smooth out details on the face.
Try doing upscaling only and let me know what happens.
Thank you Lingyu, I did try that and yes, it worked much better without the face recovery. Thank you.
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.
