I am applying Photo AI to old B&W pictures which are both grainy and sometimes out of focus. The tool does a great job identifying and recovering the faces, getting rid of the grain and bringing them into sharp focus. Unfortunately it only does this in a fuzzy rectangle around the face. So there is an island of clarity which ends in a jagged boundary at the neck and in the middle of the hair at the top of the head and around the ears. If it is going to do such a great job there, it needs to do a similar job on the rest of the photo so it looks natural.
I too have this problem which makes it unusable or takes alot of work to get something that doesn’t look funny. usually what I have to do is make two different versions. One without face recover and one with it. Then in photoshop I overlay the one with facial recovery on top of teh other and lower the opacity of it so it blends in.
I think when doing facial recover it would be great for the software to change the whole image so that it blends better
I see this problem as well. Restoring photos with people from the 80’s and 90’s when folks had bigger hair than they do now (see screen shot), maybe they should put a masking feature in for us to help the software to decide what it is recovering?
Any response yet from Topaz about this issue?
I have the same kind of problem. Although the face is recognised and Topaz does a good job of recovery on most of the face it leaves an area untouched. I need to be able to mask the area of the face that needs recovery.
Have similar problems with “Adjust lighting” and/or “Balance color” enhancements when used in conjunction with “Face recovery.” I frequently (more often than not) have to run the image through twice as you mentioned - the first time to adjust lighting or color and the second time to recover faces. otherwise, the face recovery does not match the light or color enhanced and faces wind up inside boxes with clearly different tones/lighting than the rest of the image. Reordering the enhancements sometimes works, but usually have to run the image through twice.
