I just tried your suggestions. By adjusting the lighting did help with the shadows, but of course blew out the highlights.
For Recover Faces, turning off hair and neck (two of the three features I want) is a loss for me. It would be better for you to say TAI is just not able to do that yet. TAI did such a fantastic job on the hair parts it rendered it’s a shame to lose that. There is just as much detail in some other parts of the hair as the bangs. Still, I did it as you suggested.
I applied denoise and pushed the Original Detail slider up, and applied sharpen to the subject.
So now I have a sharp face and blurry hair. That’s not really what I wanted, but OK. It’s not usable because that creates an odd clarity juxtaposition.
I’m pretty sure this has nothing to do with not having enough data. Straight line cut offs at an exact right angle suggest this is more like a limitation on the distance from the face. TAI creates a box with the face in the center and then renders only what is in it. Anything outside of that will not.
To add support to that theory, TAI correctly rendered a portion of the blurry part of the hair. See the circled area in the second photo right at the corner of the rectangle. TAI was doing just fine until it reached the programatic boundary of the box.
Also, half of the ear is rendered and the other half is not. There is plenty of data on that ear.
Here’s more of the same image where you can see it is doing the same thing to the baby.
This will occur on any big updos and any neck areas as well that have neacklines different then t-shirts. As mentioned, the Ideas forum has a feature request to expand the area manually for cases like these, on which you can add your vote! For intricate recovery cases like this, you can take parts you like and uses layers in Photoshop to keep the best parts you want.
Here’s an example from GPAI that shows the square edges of the face recovery area pretty well.
See Middle Image
As shown in the link the only solution is to do a layer composition in PhotoShop to blend the perimeter better.
1 Like
@andymagee-52287 thanks! make sure to add your example and vote to the the feature request as well
The hair/neck issues on some areas is definitely something the development team is already aware of. Adding votes and examples on the correct thread can help the development team have access to multiple examples and showing the continuous interest for this feature as well.
That’s a good suggestion. I’ll give it some consideration. Nice work on that photo of your wife BTW. You did a great job.
GPAI Redefine looks like a good product. Unfortunately, I only have PAI and I’m only impressed sometimes. It’s difficult for me to dump more money into a company that is currently letting me down just so I can correct the problems that PAI made or could correct but has been limited.
My experience here has not been a good one. It’s like the employees don’t even use their own product and are using all of us a beta testers. They want us to buy and then don’t give help. They don’t even know their own product. We have to write the bugs. We have to give product suggestions. The fastest way to improve your software is to use it yourself. I don’t like that they want me to do their job. I choose not to.
This does not make sense to me.
Big updos? What about the baby?
I wanted to mention that this issue can occur with larger hair areas, and the development team is aware of it. The enhancement you’re referring to is called Recover Faces, and we do have the option to disable hair/neck processing if the results aren’t as expected or if harsh lines appear. While Photo AI is optimized for digital photography and can definitely help recover older files like yours, other areas such as the hair and body may still need additional work to achieve a smoother blend. Unfortunately, Photo AI isn’t capable of recovering the remainder of the file at the moment, especially within the Recover Faces tool. However, the team is continually working on tool updates and new features, so this may improve in the future!
There’s already an ongoing thread discussing expanding the processing area for Recover Faces and the development team is aware of it popularity. Whether or not you want to contribute your examples to that thread is completely up to you, as it’s a community-driven vote, but I just wanted to make sure you were aware of it. It’s not mandatory, but it could be helpful.
At Support, we use the app frequently, and I’ve shared a few workarounds that may help you get the best results. Ultimately, it’s about finding what works best for your needs and workflow. For instance, Andy mentioned using Photoshop to blend the edges more smoothly. In this thread, he said, “As shown in the link the only solution is to do a layer composition in PhotoShop to blend the perimeter better.” - This method is quite common, especially for recovering older photos. Andy uses three programs to achieve his results: Photo AI, Photoshop, and Gigapixel. Gigapixel can be used for file recovery and upscaling, as it also has generative models to add details, similar to what he demonstrates in his example, in his last file.
Happy editing, and feel free to reach out if you have any questions!
The black text on black background in the warning messages is now fixed in version 3.6.1 
Releases Notes
1 Like
