Which Face Enhancer do you use?

[TheBigSmokey]

I'm pretty new to this. Wondering what people prefer.

[mrscorchedsilk]

Codeformer does much better with enhancing the teeth & eyebrows. All the other ones aren't as good for me.

[mrscorchedsilk]

When I use the option "none".

Does that mean it is using a face engine called "none"

Or does it mean that it will not use any face engine on it?

I am asking this because it takes a much longer time to export a video with the 'none' option compared to just using the original roop.

Thanks

[Siraj-HM]

Yes I noticed this too. And none takes quite some time to process as well.

[vin901]

None is fastest for me, there was a post that said it was a bug, but if you update to latest as of 8/4/23 doesn't seem to be an issue any more.

Codeformer gave me artifacts when the face is looking away from screen, and Gfpgan looked bit better for me I think.

But it might vary depending on the clip, haven't done many.

[mammuthus]

Visual

• None: This obviously depends entirely on the source and can appear blurry and not detailed.
• Codeformer: It's better at masking minor tone/color imperfections (such as eye bags), but jitters are noticeable.
• DMDNet: Jitters are less noticeable compared to Codeformer, but it doesn't hide tone/color defects as well. Occasionally, I notice some over-sharpening in the output video, especially on bright areas like eye glares. Teeth also appear very bright, and faces seem overlit.
• GFPGAN: This offers the best compromise, providing the best smoothing and face blur.

Performance tests

None - 20.2 fps
Codeformer - 1.85 fps
DMDNet - 9.6 fps
GFPGAN - 5.3 fps