Many Image referece for one target video.

C0untFloyd / roop-unleashed

Evolved Fork of roop with Web Server and lots of additions

GNU Affero General Public License v3.0

2.23k stars 519 forks source link

Many Image referece for one target video. #59

Closed tikendraw closed 1 year ago

tikendraw commented 1 year ago

Instead of using one image for the entire video and wrapping it, take many source images for one target video to match the facial structure better.

phineas-pta commented 1 year ago

impossible with currently used model (insightface inswapper)

tikendraw commented 1 year ago

Understandable. 👍

phineas-pta commented 1 year ago

this altermative to inswapper may be what u want https://github.com/eric-ai-lab/photoswap

C0untFloyd commented 1 year ago

Nice find but this looks more like a Diffusion Lora paired with Control.NET. Also it seems you need to train a checkpoint. This is too much for a quick deepfake I think.