[FACE SELECT AREA] - Possibility to refine the selection of faces using a brush or selection square

Now I’m testing PhotoAI on old scanned prints: face recognition is generally ok and the result is often excellent. Several times however parts of the hair are not included in the face as well as part of the neck. It would be nice to be able to refine the faces with a brush as you do for the selection to be sharpened.
Two examples
Screenshot_20230209_120135

Screenshot_20230209_120001

Yes, would be good if - as part of the features of a maual vs AI Masking Brush - hair, neck, ears or any desired head parts could be manually selected to be sharpened & a pixel range transition to softening could be specified (real-time preview visible slider).

1 Like

As was discussed before, when faces are enhanced, the neck is usually left a mess.
A couple ideas on correcting that deficiency;

  1. If possible extend the facial recognition to include area below face that has same or similar skin tone/color. If it can smooth the face it can smooth the neck.
  2. Add the option to physically select the face area so the user can include the neck area. This would allow for different neck lines created by different clothing types.
  3. Perhaps include a body recognition, based on skin color/tone matching face color/tone. This could prevent arms/legs from being left a noisy mess after the face has been corrected. If not auto recognition, at least an area selection for clean up. Use a tool to outline the arms/legs and such for clean up. Maybe at a later date include recognition and a way to clean up fingers and toes without turning them into eagle talons.

Very often the app does not recognize a face when its very obvious… so perhaps just allow to drag a box around each manually would help… or even better allow to specify eyes, mouth, nose, ears, etc… sometimes to specify a side profile as well when a full face is not shown

Thanks for the idea and suggestion! Make sure that you do not forget to vote for your idea :slight_smile:

How to vote? seems only logical to train the AI where the face items are

You can vote for your idea at the top of this Forum thread by clicking the Vote button :slight_smile:

is there a way to change the size of the face window, sometimes it does not cover the whole face??

This is not currently a feature in Topaz Photo AI; however, we appreciate the idea and suggestion! Please make sure that you do not forget to vote for your idea :slight_smile:

I’m trying to restore a 1940’s portrait that has a photo texture to it. The face recovery looks beautiful on the woman’s face but does not include all of her hair or her neck where the texture remains highly visible. I’d like to be able to enlarge the area that is recovered under face recovery. Would be great if I could expand the face selection square or draw in the area with a masking brush.

2 Likes

Try this:

I use this regularly. It won’t be easy, it’s worth watching a few videos about it.

2 Likes

We’re working on improving our Face Recovery models to blend better with the area immediately around the face. In the meantime, try using the Upscale filter to enhance the resolution of the entire image. There isn’t much we can do about the texture seen in the scan since it isn’t necessarily a digital imperfection.

2 Likes

Thanks for your response, I’m not looking to remove all of the texture, but if it can as you can see it removed it from the half of the face that has been processed, I see no reason why I can’t improve the hair and her neck if I could just enlarge the area that facial recognition sees. Brush it in as though making a mask layer.

1 Like

I see what you mean, but the face detection algorithm doesn’t quite work like that for the rest of the image unfortunately. Try actually turning down the strength of the slider and upscaling so there isn’t as stark of a contrast. However, it won’t get a perfect result due to the systemic limitations of the software.

Thanks. It’s more just messing around with a friend’s mother on Facebook. I love what it accomplished with her actual face. So maybe my request is for some future method of dealing with textured photos and also with newsprint dot patterns while I’m at it!

1 Like

Got it, that could be something we explore down the line but our current focus is on digital images rather than scans!

1 Like

So i’ve seen some great ideas to add to the face recovery feature and I wanted to also add this, the ability to choose where the recovery is used. I’ve seen my fair amount of photos where the image is quite blurry and then you see these perfectly sharp and restored boxes around the face, a clear sign of upscaling when the idea is it to look as natural as possible. Sometimes I’m restoring an image where the hair is completely blurred and a small boxed portion of the hair is now sharp. I would love the ability to select how much of the hair and face I want restored, on top of all the other things people have talked about, like animal face recognition, gender/age specific tools when it comes to the amount of smoothing used, etc. Sometimes I’m satisfied with the face but feel the lack of restored hair ruins it. Maybe we can have a specific hair recovery tool?

When using the facial recovery process, it frequently does not cover things like a neckline into the clothing. This leaves a very obvious quality difference in the skin which can be rather unpleasant.

I have also thought about this, retouching software such as PortraitPro Studio Max for example allow adjusting, it would be nice to have this.

1 Like

We’re working on improving the Face Recovery models to not have such a stark contrast with the rest of the body. In the meantime, try turning down the strength of the model and turn on/up the Upscale model which will enhance the resolution of the entire image.