Recover Faces

Clarity in facial features enhances an image’s communicative value. The Face Recovery enhancement recovers small or low-resolution faces by generating a high-quality version and blending it with the current face.

Important: The AI model will need key face details to detect the face and be able to be applied. If the AI model does not detect the face it is not possible to force it.

Versions

Recover Faces v3 is the newest model available, and was released in v1.6.0. The new v3 model is a generative recover faces model that processes each face separately. Each additional face selected will increase processing time. For example, 4 similarly sized faces will take 4x the time compared to 1 face of the same size.

For faces already with decent quality and below the 512x512 pixel limit, Recover Faces v2 will still do the job and process faster. For large faces that need recovery, blurry faces, or anything degraded enough that Recover Faces 2 struggled, 3 is the better starting point. and the quality difference is worth the extra processing time.

Availability

Recover Faces 3 runs on NeuroServer and processes locally on NVIDIA, AMD, and Apple Silicon (macOS 14 or later). Cloud render is available for unsupported hardware, up to a maximum output size of 100MP.

System Requirements for Local processing:

Windows: NVIDIA or AMD Graphic card with 8GB VRAM or more
Mac: Apple Silicon (M-series) machines with 16GB RAM minimum

Not supported on:

Intel Macs
Windows ARM / Snapdragon systems

If your machine does not meet the requirements for local processing, you can still use the Cloud processing option.

How to use

On file import, Topaz Photo will analyze the file and look for faces. Then, you can either manually add the "Recover Faces" enhancement or if you have Autopilot on, it will kick in automatically when faces are detected.

Once the Recover Faces enhancement is added, make sure that all the faces you want to enhance have a yellow selection around their faces.

Important: The yellow selection rectangle is not showing what area is affected by the enhancement. The yellow rectangle is only to determine which faces will be affected by the enhancement (yellow rectangle), and which will not (grey rectangle). In no way is this representative of the affected area.

Use the Controls tab to change the Strength for Recover Faces.
To get a good blend, check to see if Hair and Neck should be toggled on or off. You can toggle this on/off in your Selection Menu of Recover Faces or in your Edit (Topaz Photo menu on Mac) > Preferences > Recover Faces > Parts to enhance.

To get the best blend possible, make sure to not only play with the Hair and Neck options, but make sure to recover the rest of the file, with other enhancements. Reach out to support@topazlabs.com with your original and current results attached to your email and we can return steps on how to get the best possible results.

If a face is not "detected" and not available to be selected in the Selection options of Recover faces, it is not possible to "force" Topaz Photo to "detect" this face. This can occur if the face is too small, does not have enough details to be recovered, if the face has hidden facial features, is sideways, etc.

When to Use

Low Resolution Faces

Use the Face Recovery filter on low resolution faces, even if it is large and takes up most of the image. Resolution is measured in pixels.

Small Faces

Use the Face Recovery filter when the faces are small and far away from the camera. The model generates the best results when the face still has sufficient detail. See the full image and zoomed-in portion below.

When Not to Use

Portraits

Using the Face Recovery filter can over-process faces that are already high resolution. This causes a loss of detail and creates a “plastic” feeling because it is too smooth. Make sure to look at the Sharpen Portrait enhancement instead.

Unrecognizable Faces

Do not use the filter for out-of-focus images, very low-resolution faces, or on faces that lack key details. The lack of detail hinders accurate facial reconstruction and will generate unrealistic artifacts.

Support

If you get any unwanted results or issues, send in the Original file and current results you are getting to support@topazlabs.com and we can provide steps to get the best possible recovery for your file.

Spot Healing

Preserve Text