have the same issue with pictures featuring redhead person (same with gigapixel). it would need a special analysis, and when the Ai see a redhead person, it turn off face recovery and noise reduction.
In this extreme example, PAI does not even find a face in Autopilot:
In manual mode, it does nothing to the face at all:
Here is the original, treatment with GFP-GAN, and finally GPAI, which DID process the face (please view at 100%, even though the forum is limiting full size display [original is 6731px wide]):
it seems that if the software can’t see a face, this feature is totally turned off.
Even turned on manually it couldn’t find the face, but GPAI can.
i know, that’s not what i told. i was telling that if the software doesn’t see the face, the face recovery can’t apply. and yes i understood that Gpixel saw the picture, i noticed the same amount of error. sometimes photo Ai see the faces, sometimes not, i got “no face”, on big face picture take half of the picture, so pretty hard to miss it lol 
. certainly this point must get a bit better ;).
Thanks, though I was hoping that a manual override might help it along…
So it seems PAI must be further tuned for faces. Work in progress!
1 Like
which kind of makes sense, to enhance a face you need to be able to locate it first, so the off switch can stop if processing detected faces but it if can’t find a face in the first place switching it on won’t help.
absolutly, and unless the software can have a 90% positive result on picture, a manual indication of where is/are the face(s) could not be a too much bad idea. but certainly this will work better with time, gigapixel work better on this, and the software certainly share a lot of common stuffs (code, etc…)
1 Like
I have done some more test - I reverted back to 0.2, and it appears, that there is issue with my laptop only - on my PC everything was working fine on 0.2 (didn’t test 0.4) - on my laptop also face recovery set to on turns off all filters. It is kind of strange:/
Could also be fun deliberately using it where there isn’t a face, a bit like the AI ‘dreams’ eye’s and dogs stuff, that was all the rage a few years ago. Not exactly what the App’s designed for 
1 Like
Hi, Just like to point out we do not all take images with human faces in them, in Sharpen stand alone it can detect a bird from within an image so how about giving that option within both Denoise stand alone and this new Photo AI so we can work on that part of the image that is most important?
Also give the option to remove the face detection part of the image processing as for wildlife it’s just a waste of the programs efforts, Lastly I no longer use Adobe Lightroom, moved over to On1 Photo Raw 2022 so would like to see an option to open Topaz Photo from within On1 if possible. Thankyou.
“low quality” is just dictated by the size of the face, currently. Sufficiently large faces are assumed to be processed just fine by the other filters.
1 Like
I had exactly the same question like Alain.neuens with birds above water. I always select the bird to prevent the problem shown here.
If anyone had a problem with not working Face Recovery - I checked log files, it appears that some models could not be open, I spot that models in C:\ProgramData Topaz folder were from different dates, and maybe there new models could not be replaced during update(I had to rise privileges to admin when I wanted to delete them). I cleaned all models, then I reinstalled 0.2, face recovery started to work. I will try now 0.4, overall in my case I admit to clean models folder before installation of new version:)
Hi, Autopilot worked OK BUT when saving back to Lightroom it said the module was missing (Had Denoised only) and unable to return image. Tried on several images (all DNG). Is there an image size limit please? I like the concept of an all in one program.
Kind regards
At the moment DNG images are not supported to be written out as they are not mentioned in the preferences, remember it is a early release so not fully functional. In the meantime use TIF/TIFF.
Thanks for that. Tried exporting from lightroomas a Tiff (16bit, uncompressed) and same problem. Will save image outside the LR / Photo AI continuum as a TIFF and then open it and see what eventuates.
Noticed a couple of problems immediately, but some are likely temporary issues at this stage of development.
-
Window size and position is not remembered between sessions and cannot be set by dragging to a Microsoft PowerToys FancyZones target for quick resizing to useful proportions. PowerToys problem might indicate an interface design choice that somehow bypasses standard Windows behaviours (Gigapixel AI, Sharpen AI and DeNoise AI all have FancyZones compatible Window behaviour, although Gigapixel AI alone does not remember size and position).
-
Installing Photo AI does something to corrupt system wide file associations for images. Now applications such as Microsoft Edge no longer save PNG files as “Portable Network Graphics” file type but save them as “Topaz Photo AI 0.4.0” file type instead (PNG). The files still open in the correct associated programs, so rather than having been conventionally “stolen” by Photo AI, this seems not to be a straightforward file association problem. I have not found time to track down the registry entries on this.
-
There is a bit of uncanny valley with familiar faces (such as my own) due to changes in eye shape (inner corner of eye curving up converted to inner corner curving down). Might just be the nature and limitations of AI, but I don’t recall seeing this happen in other Topaz products so far. I will compare same photo on other Topaz apps and upload sample later.
Quick feedback on a few test images I was able to manually recover quite nicely (very old color photos) using PS:
Photo AI is SOMETIMES doing a good job. Face recovery is mostly useless for me as it changes the appearance of the face (basically the eye shapes get distorted, all eyes are resized to smaller/more “Wester Europe” style, corners of mouth get biologically wrongly reshaped). Another big problem with face-improvements is the very visible seam between where the face enhancer works (removing all detail) and where it does not (lower part of neck, lots of detail remain) and the resulting VERY different look of the faces (plastic/rubber doll appearance after “enhancement”) versus the rest of the photo (grainy, naturalistic look).
Oversharpened edges, halos, colors being “extremified” …
It’s a beta. I’ll keep watching.
1 Like
I am very happy with Photo AI; the results are generally good and saves a lot of workflow time. I realize it is a work in progress. I have a question, though:
On running with Autopilot, noise is dealt with adequately, but if you then resize the photo, even using modest (1.4) settings, the image becomes noisy again, requiring further processing in DeNoise AI.
Bug? Or did I not read the manual thoroughly?
