As others have said and I’ve written here in the past as well: why does face detection make everything blurring?
it does not seem to matter type of subject, type of image (photo or rendered/art). I’ve yet to really see anywhere where this options as made an improvement
Original is on left and then top /right face detection on/auto
Lower Right, Face Detect Off.
To me… the lower right looks extremely detailed, not blurry, eyelashes, eyebrows well defined (honestly incredible what this app can do, with such a pixelated mess on the left!).
But the Face Detect version looks… terrible 
Face detection is not really intended for large faces that take up a significant portion of the overall image. That’s why the result bottom right is better.
Well that was small crop of larger image. I was a full body image. Feet to above head.
Can you upload the original so I can run some tests on my end and make some workflow recommendations to get the best results?
I immediately turn the facial recognition down to 25% from the set 80%, and sometimes I go lower or turn it off altogether. It often leaves faces looking plastic by taking out every contour and blemish. It needs some serious fine-tuning.
The autopilot needs to ignore large faces (except when they are of very poor quality).
Thanks for this input.
One thing I want to clarify is that I see comments like “big faces taking up most of photo” aren’t suitable for face detection.
Well, from what I see it’s any kind of face suffers. 1/10th of image, 1/2 of image. Doesnt’ matter.
The most obvious things I see happening are in eyelashes, eyebrows and the detail in lips. With face Detect on, they all get smudgy looking, for lack of better way to describe it.
I keep it on and try to look around the face… did it fix something I’m missing? do the eyes, or something I’m not paying attention get improved (at the cost of the blurriness)… ?
And wow, I can’t see it. Nothing looks better, image after image.
Turn if off, and all these areas become crystal clear (or as clear as physically possible given the understood limitations).
So I really am trying to find a single image out of thousands I’ve processed where Face Detection has actually made anything better, clarified detail… and I haven’t come across a single photo or generated piece of artwork.
There’s an app for Android (prob ios too) called faceTune. Oddly enough THIS app can correct small issues with faces, like weird eyes and other minor defects. It does not enhance the image really but it can take a person who has 1 eye 1/2 blinked… and kind of correct it. Same with some AI art renders. It can actually fix a few of the weird noses/faces that can otherwise kill a great AI rendered image.
I’m not trying to compare the apps. They do totally different things. But when we mention the word “Face detect” one actually does it… But seems to me that Topaz is … doing… absolutely nothing… that I can tell, except make things blurry.
Topaz Photo needs to re-examine what the whole purpose of face detect is. I don’t see any value to it as it currently is implemented.
Yes, Topaz AI is INCREDIBLE in how it can increase the detail of a low res image. But the face detect?
I would revisit why this is even in the app until it can actually provide useful output.
If the face in the original image has sufficient detail, then the regular upscale without face recovery is almost always going to be better. That’s always been the case since the feature was first introduced. Basically, if you can see lip detail, eye detail, etc in the original image, then don’t use the face recovery.
The main limitation is that there is a pixel limit to the face generated, and it isn’t very high. The face recovery option was first created to solve the issue with background faces that could end up looking like the living dead after an upscale, not the main subject.
There were alot of forum complaints about this, and if the feature were removed, they would soon return.
@eric.holbrook @alanhusband-677420 Thanks for all your insights. We’re releasing new and improved Face Recovery models based on your feedback with v2.0 on September 7!
Please update when that’s available and let us know what you think!
I’m guessing that all the faces you’ve tried are medium to large in this context, since Face Recovery is definitely better than the normal upscale for small faces (and probably faces of very poor quality too). I think the crossover happens when the face is about 100 pixels high, although I haven’t done much testing. I also use the 4x upscale, which may change matters compared to 2x.
Take one of your images, downscale it so that the faces are 50 pixels high or smaller, and compare the upscales with and without Face Recovery.
But there has nothing changed for the Face-Recovery-Model in PhotoAI 2.0… 
I still have Problems with FaceRecovery and a simply Selection-Tool would really help! 
1.)
Some Hair-Areas are not detected and are still unsharp…
Could you maybe implement a “Quick Select”-Tool for the FaceDetection to manually select Areas, which are not automatically detected (like in the “Preserving Text”-Tool)?
2.)
There are some Cases where still strange Artifacts appear on the Face, Teeth, Nose etc…
I think I’ve come to realize there’s flaw in perhaps the app and my own thinking.
Recover faces should not be used if the face is not even remotely blurry. If the face is basically clear already but say, low-res, it does far more harm than good
But the problem here is, the app in auto detect mode, always turns on Recover Faces. So this is partly to blame for the app just assuming everything needs Recover Faces.
The reality is, if your photos (or art or whatever) has clear faces, just perhaps on the low res side of things… DO NOT use Recover Faces. It offers absolutely no benefits.
I think this feature is a hold over from the “blurry photo, let’s clear up photos from 1950 of my parents” craze… and really probably what got this app developed in the first place.
But I think nowadays, as implemented, it has little value.
I would far rather see a true “Recover Face” feature along the lines of Some IOS and Android apps that truly do “Recover” faces. Faces where someone’s eye is 1/2 closed, or nose got mushed or you want to clarify the eyes, or clean up the face basically. Recover it.
For AI Art this would be phenomenal benefit.
Right now for clear artwork, photos, etc… Recover Faces is a wasted feature that does not benefit anyone except to photos taken during WW2.
I challenge anyone to show me a modern photo example where it has improved a photo. I’ve definitely seen it happen! Don’t get me wrong. It was a photo of my dad from the 1960’s. And it was blurry yeah… did pretty well although ended up looking like no one we knew.
But this was an edge case I think. This feature as created, is not useful on day to day photos.
Here is a full size image
Orig is 1024x1536
Now compare the upscale + Recover Faces Off (top) vs Recover Face Off (bottom)
The version with Recover is blurry, loses detail, etc… But it’s what the AI Autopilot has chosen
But the problem I guess is that Auto Pilot still wants to try to fix. It detected a face, and it tries to correct, and fails. The face does not need (this) kind of correction.
This is a 200x zoom of
the origin pixels
Face Recover On (auto all)
Face Recovery Off
Original
Face Recovery
FR off
There’s def a part of me that does NOT like the no FR version either. Looks very flat and lacking all texture, realism… vs the perceived original… that looks… not like a de-noised Photoshop thing.
But The FR version looks terrible too but seems to try to add some texture back into the mix.
But there’s really no way to say the 200x FR version is something you’d buy software to generate THAT version vs the original
yes I have it set that way. But does make me wonder. There are so many unused features for me it seems
Upscale
Remove Noise
Sharpen
Recover Face
Preserve Text
The ONLY one I really only can ever use consistently is the upscale. The Remove Noise and Sharp are smart enough to shut off if not needed. I do find weird use cases where I can use them but rare.
The Recover Faces is always on, whether needed or not.
You can turn this off in the Autopilot preferences or turn down the strength of the slider.
