search

C0untFloyd / roop-unleashed

Evolved Fork of roop with Web Server and lots of additions
GNU Affero General Public License v3.0
2.33k stars 541 forks source link

Deepfake "remaining" face when obstacles are in the way? #950

Closed tamerkalieh closed 1 month ago

tamerkalieh commented 1 month ago

Hello,

The problem happens when using a live camera or simply using it on videos. When any object hits the face, even if it's just a tiny object that covers the face by 2-10%, a wider area around the obstacle bugs out, and it looks like two layers (old and new faces) are on top of each other.

Is there a way to fix it?

Using: roop unleashed 4.1.4 python: 3.10.14 • torch: 2.1.2+cu118 • gradio: 4.38.1 Codeformer as post-processing

A great example where I experience it is when using a video where I pet a dog and the dog licks over the face or covers it by a tiny bit.

C0untFloyd commented 1 month ago

Activate xseg masking in the faceswap tab.