Face recovery & selection

Utilize Face Recovery to improve images where facial attributes could be enhanced. Low resolution, blurry, and unclear details of human faces are best used for this feature.

• Face Recovery detects and recover human faces. It works best on blurry, low resolution, or artifacted faces, but may remove detail from high-quality faces.
• Face Selection - detects high and low confidence faces to enhance

Understand Face Recovery AI

When enabled, the AI scans across the entire image searching for the eyes, nose, and a mouth of a human face.

It determines the number of faces recognized and categorizes these faces between two types: high confidence and low confidence.

High Confidence faces follow a pattern of being the subject in focus of an image and distinguish clear details of the eyes, nose, and mouth.

Low Confidence faces are determined by their lack of detail like blurriness or out of focus which is usually found in group settings or subjects where the head is turned away.

The Face Recovery model does not use Generative Al or diffusion models. They are trained from digital images from digital cameras. Certain types of images are not always easily recognized such as old scanned photographs.

How to Use Face Recovery

In Additional Settings, click the toggle switch to activate the feature.

Click on the button Select Faces to Recover to enhance.

In the Preview, the image shows boxes around faces that the AI has detected. Click on the desired boxes that need enhancement.

A blue colored box indicates High Confidence.

A gray rectangle indicates Low Confidence.

Click on the Low Confidence face if you wish to include it in the Face Recovery enhancement. You can also click on High Confidence faces if you do not want these to be enhanced.

After faces have been selected, Click on Apply.

After the AI Model finishes enhancing, adjust Face Recovery Strength by sliding the bar to improve the face. The AI defaults a numbered value of enhancement quality. Fine tune the results by moving the slider bar to increase/decrease the intensity of the enhancement. Skin can appear smoother or clearer depending on the strength.

Photo by Pavel Danilyuk

Face Recovery Generation v2

Released as a beta in v7.4.0, this AI model now adds another layer of enhancement possibilities that v1 did not offer before.

Our models are designed to preserve the original image but since Face Recovery is specifically designed to reconstruct damaged face images, the Face Recovery model will sometimes change expressions slightly or may smooth skin, hair, or small facial details.

This version not only improves upon v1 but offers two modes:

• Realistic mode closely mirrors the original Face Recovery model but with improved quality overall. For scenarios with very low resolution or tiny faces, Creative mode might offer better results.
• Creative mode generates additional details that might be missing from the original image, producing high-quality results even from very low-resolution inputs. It’s incredible what can be achieved with just a few pixels. Note that the Creative mode does not use a diffusion-based generative model, so if you prefer traditional machine-learning models, Gen2 Face Recovery is still a great option.

Check out these two modes in action here.

The example below showcases the difference between V1 & V2 realistic and creative. Notice that certain facial features are altered differently in Creative mode.

Photo by Oskars Rotstein

Images that are most optimized with this model are:

• low resolution images
• blurred out or obscured faces
• group photos