Face Recovery

Topaz Video AI Ideas
1

I had a dream…what do you think? For all three reconstruction models Iris, Rhea and Proteus :slightly_smiling_face:

Face Recovery

Face Recovery Strength
Value 0 = models default

negative values = less face recovery strength
positive values = more face recovery strength

Face Recovery Sensitivity
Value 0 = models default

negative values: Does less face recovery on faces that are difficult to recognize and reconstruct, for example if just few pixels for facial reconstruction are available.

Particularly useful to prevent faces from being distorted when taking long-distance shots. If you set the control bar to the far left, the model will only reconstruct very good recognizable faces, which is usually the case with close-ups

positive values: more progressive to recognize faces in order to reconstruct them

2

That sounds like a great idea. While giving users the option to manually adjust settings is helpful, it would be even better if the program could automatically enhance elements based on the specific model being used.

3

A feature that uses all models simultaneously to automatically determine the optimal settings for upscaling could combine the strengths of each model, provide the best quality, and improve user convenience.

4

@ForSerious The idea is that face recovery can be adjusted by users and also turned off, so a Proteus without it would still be possible.

5

Leave Proteus out of it—since it has no special face enhancing like the other two—and this is an idea everyone has wanted at some point.

6

I cannot agree more AVClabs Video Enhancer has a slider to adjust the level of face enhancement and something similar would be very much needed in Topaz!

7

Bump this,
I consider this as “must have” feature.

I’m not a poweruser of Topaz AI, and I don’t how AI works on deep levels, but so far the videos I processed are all improvements… except faces, they are generally deteriorations.

there must be an option to control the weights for facial reconstructions as the results are pretty much always bad for me, and I tried every option there is, all models with different settings.

I’m still on v5.2 but the changelogs didn’t show anything regarding this problem so I didn’t bother updating

1 Like